更广阔的旅程，更广阔的边界

A Broader Journey with Broader Boundaries

在生成式人工智能和所有其他不断发展的人工智能系统激增的世界中，构建产品需要关注尽可能广泛的数据池视图、可以实现的旅程以及它们带来的风险。

In a world where generative AI and all the other evolving AI systems proliferate, building an offering requires focusing on the broadest possible view of your pool of data, of the journeys you can enable, and the risks they raise.

笔记

NOTES

1. 1 . David C. Edelman 和 Mark Abraham，“人工智能时代的客户体验”，《哈佛商业评论》，2022 年 3 月至 4 月， https:// hbr .org /2022 /03 /customer -experience -in -the -age -of -人工智能。

2. 1. David C. Edelman and Mark Abraham, “Customer Experience in the Age of AI,” Harvard Business Review, March–April 2022, https://hbr.org/2022/03/customer-experience-in-the-age-of-ai.

3. 2 . Dharmesh Shah，“向ChatSpot.ai打个招呼：一体式 AI 支持的聊天应用程序，让您成长得更好”，YouTube 视频，2023 年 3 月 6 日， https : //www.youtube.com/watch ?v = fayBEXSKyoI &t = 5秒。

4. 2. Dharmesh Shah, “Say Hi to ChatSpot.ai: The All-in-One A.I. Powered Chat App for Growing Better,” YouTube video, March 6, 2023, https://www.youtube.com/watch?v=fayBEXSKyoI&t=5s.

改编自 hbr.org 上发布的内容，2023 年 4 月 12 日（产品#H07KSV）。

Adapted from content posted on hbr.org, April 12, 2023 (product #H07KSV).

网络效应对人工智能的作用有所不同

Network Effects Work Differently for AI

AI的价值在于准确的预测和建议。但与依赖将供应（如电力或人力资本）转化为产出（如照明或税务建议）的传统产品和服务不同，人工智能需要大量数据集，这些数据集必须通过来回的客户互动保持最新。为了保持竞争力，人工智能运营商必须收集数据、分析数据、提供预测，然后寻求反馈以完善后续建议。系统的价值取决于用户提供的数据，并且随着数据的增加而增加。

AI’s value lies in accurate predictions and suggestions. But unlike traditional products and services, which rely on turning supplies (like electricity or human capital) into outputs (like light or tax advice), AI requires large datasets that must be kept fresh through back-and-forth customer interactions. To remain competitive, an AI operator must corral data, analyze it, offer predictions, and then seek feedback to sharpen subsequent suggestions. The value of the system depends on—and increases with—data that arrives from users.

该技术的性能（准确预测和建议的能力）取决于称为数据网络效应的经济原理（有些人更喜欢数据驱动的学习）。这些与熟悉的直接网络效应不同，例如随着用户的增长而使电话变得更有价值的效应（因为您可以呼叫的人更多）。它们也不同于间接或二阶网络效应，间接或二阶网络效应描述了越来越多的买家如何邀请更多卖家加入平台，反之亦然——当更多卖家出现时，在 Etsy 上购物或在 Airbnb 上预订变得更具吸引力。

The technology’s performance—its ability to accurately predict and suggest—hinges on an economic principle called data network effects (some prefer data-driven learning). These are distinct from the familiar direct network effects, like those that make a telephone more valuable as subscribers grow (because there are more people you can call). They are also different from indirect or second-order network effects, which describe how a growing number of buyers invites more sellers to a platform and vice versa—shopping on Etsy or booking on Airbnb becomes more attractive when more sellers are present.

数据网络效应是一种新的形式：就像越熟悉的效应一样，用户越多，就越有价值科技群岛。但在这里，价值并不来自同行的数量（如电话）或许多买家和卖家的存在（如 Etsy 等平台）。相反，其影响源于技术的本质：人工智能通过强化学习（预测和反馈）来改进。随着智能程度的提高，系统可以做出更好的预测，增强其实用性，吸引新用户并保留现有用户。更多的用户意味着更多的响应，从而进一步提高预测的准确性，从而形成良性循环。

Data network effects are a new form: Like the more familiar effects, the more users, the more valuable the technology is. But here, the value comes not from the number of peers (as with the telephone) or the presence of many buyers and sellers (as on platforms like Etsy). Rather, the effects stem from the nature of the technology: AI improves through reinforcement learning—predictions followed by feedback. As its intelligence increases, the system makes better predictions, enhancing its usefulness, attracting new users and retaining existing ones. More users mean more responses, which further prediction accuracy, creating a virtuous cycle.

以谷歌地图为例。它使用人工智能来推荐到达目的地的最快路线。这种能力取决于预测替代路径中的流量模式，这是通过利用来自许多用户的数据来实现的。使用该应用程序的人越多，它积累的历史数据和并发数据就越多。借助大量数据，谷歌可以将无数预测与实际结果进行比较：您是否在应用程序预测的时间到达？为了完善预测，该应用程序还需要您的印象：说明有多好？随着客观事实和主观评论的积累，网络效应就会开始发挥作用。这些效应可以改善预测并提升应用程序对用户和 Google 的价值。

Take, for example, Google Maps. It uses AI to recommend the fastest route to your destination. This ability hinges on anticipating the traffic patterns in alternative paths, which it does by drawing on data that arrives from many users. The more people use the app, the more historical and concurrent data it accumulates. With piles of data, Google can compare myriad predictions to actual outcomes: Did you arrive at the time predicted by the app? To perfect the predictions, the app also needs your impressions: How good were the instructions? As objective facts and subjective reviews accumulate, network effects kick in. These effects improve predictions and elevate the app’s value for users—and for Google.

一旦我们了解了网络效应如何驱动人工智能，我们就可以想象该技术所需的新策略。

Once we understand how network effects drive AI, we can imagine the new strategies the technology requires.

OpenAI 和微软

OpenAI and Microsoft

我们先从 OpenAI 和微软的联姻说起。当我们对 ChatGPT 进行 Beta 测试时，我们对其创造性、人性化的响应印象深刻，但也意识到它陷入了困境。它依赖于 2021 年最后一次收集的大量数据，但缺少近期事件和当前天气等信息。更糟糕的是，它缺乏强大的反馈循环：当建议是幻觉时，你无法敲响警钟（该公司确实允许“拇指朝下”的回应）。然而，通过与微软的联系，OpenAI 找到了一种测试预测的方法。 Bing 用户提出的问题以及他们如何评价答案对于更新和改进 ChatGPT 至关重要。我们想象，下一步是微软为其算法提供其维护的大量用户数据云。当 ChatGPT 消化无数的 Excel 工作表、PowerPoint 演示文稿、Word 文档和 LinkedIn 简历时，它会更好地重新创建它们，让办公室居民感到高兴（或害怕）。

Let’s start with the marriage of OpenAI and Microsoft. When we beta-tested ChatGPT, we were impressed with its creative, humanlike responses, but recognized it was stuck. It relied on a bunch of data last collected in 2021, but was missing information such as recent events and the current weather. Even worse, it lacked a robust feedback loop: You couldn’t ring the alarm bell when suggestions were hallucinatory (the company did allow a “thumbs down” response). Yet by linking to Microsoft, OpenAI found a way to test the predictions. What Bing users ask—and how they rate the answers—is crucial to updating and improving ChatGPT. The next step, we imagine, is Microsoft feeding the algorithm with the vast cloud of user data it maintains. As it digests untold numbers of Excel sheets, PowerPoint presentations, Word documents, and LinkedIn résumés, ChatGPT will get better at recreating them, to the joy (or horror) of office dwellers.

这里至少有三个广泛的教训。

There are at least three broad lessons here.

• 首先，反馈至关重要。为了保持智能，算法需要当前用户选择的数据流和过去建议的评级。没有反馈，即使是最好的算法也不会长期保持智能。正如 OpenAI 意识到的那样，即使是最复杂的模型也需要链接到不断流动的数据源。人工智能企业家应该记住这一点。
• First, feedback is crucial. To remain intelligent, an algorithm needs a data stream of current user choices and rating of past suggestions. Without feedback, even the best algorithm won’t remain smart for long. As OpenAI realized, even the most sophisticated models need to be linked to ever-flowing data sources. AI entrepreneurs should remember this.
• 其次，高管们应该定期进行细致的信息收集，以最大限度地发挥这些效应的好处。他们应该遍历典型的财务和运营记录。有用的数据随处可见——公司内部和外部。数据可能来自与买家、供应商和同事的互动。例如，零售商可以跟踪消费者查看的内容、他们在购物车中放入的内容以及他们最终支付的费用。累积起来，这些微小的细节可以极大地提高人工智能系统的预测能力。即使是不常见的数据位，包括公司控制范围之外的数据位，也可能值得收集。天气数据可帮助 Google 地图预测交通情况。跟踪招聘人员用来搜索简历的关键词可以帮助LinkedIn为求职者提供成功秘诀。
• Second, executives should routinize meticulous gathering of information to maximize the benefits of these effects. They ought to traverse the typical financial and operational records. Useful bits of data can be found everywhere—inside and outside the corporation. Data may come from interactions with buyers, suppliers, and coworkers. A retailer, for example, could track what consumers looked at, what they placed in their cart, and what they ultimately paid for. Cumulatively, these minute details can vastly improve the predictions of an AI system. Even infrequent data bits, including those outside the company’s control, might be worth collecting. Weather data helps Google Maps predict traffic. Tracking the keywords recruiters use to search résumés can help LinkedIn offer winning tips for job seekers.
• 最后，每个人都应该考虑他们共享的数据，无论是有意还是无意。事实和反馈对于做出更好的预测至关重要。但是您的数据的价值可能会被其他人获取。高管们应该考虑谁的人工智能将从他们共享（或允许访问）的数据中受益。有时，他们应该限制共享。例如，当 Uber 司机使用 Waze 应用程序进行导航时，他们会帮助该应用程序的所有者 Google 估计乘车行程的频率和长度。当谷歌考虑运营自动驾驶出租车时，这些数据可能非常宝贵。当像阿迪达斯这样的品牌在亚马逊上销售时，零售巨头可以估算跨品牌（与耐克相比）和类别（鞋子）的需求以及买家的价格敏感度。结果可以提供给竞争对手，或者有利于亚马逊的自有品牌产品。为了解决这个问题，高管可以避开平台中介机构或第三方。他们可以协商数据访问。他们可以努力与客户保持直接联系。有时，最好的解决方案可能是数据所有者在数据交换中进行捆绑和共享，就像银行在建立共享信用数据的方法时所做的那样。
• Finally, everyone should consider the data they share, intentionally or not. Facts and feedback are essential for building better predictions. But the value of your data can be captured by someone else. Executives should consider whose AI stands to benefit from the data they share (or allow access to). Sometimes, they should limit sharing. For instance, when Uber drivers navigate with the Waze app, they help Google, the app’s owner, to estimate the frequency and length of ride-hailing trips. As Google considers operating autonomous taxis, such data could be invaluable. When a brand like Adidas sells on Amazon, it allows the retail behemoth to estimate demand across brands (comparing with Nike) and categories (shoes) plus the price sensitivity of buyers. The results could be fed to a competitor—or benefit Amazon’s private label offerings. To counter that, executives can sidestep platform intermediaries or third parties. They can negotiate data access. They can strive to maintain direct contact with customers. Sometimes, the best solution may be for data owners to band and share in a data exchange, like banks did when establishing ways to share data on creditworthiness.

当您考虑人工智能网络效应时，您可以更好地了解该技术的未来。您还可以看到这些效应与其他网络效应一样如何使富人变得更加富有。人工智能背后的动力意味着早期推动者可能会获得丰厚的回报，而追随者无论速度有多快，都可能会被留在场外。这还意味着，当一个人能够使用人工智能算法和数据流时，优势会随着时间的推移而积累，并且不会轻易被超越。对于高管、企业家、政策制定者和其他所有人来说，人工智能最好的（和最坏的）尚未到来。

When you consider AI network effects, you can better understand the technology’s future. You can also see how these effects, like other network effects, tend to make the rich even richer. The dynamics behind AI mean that early movers may be rewarded handsomely and followers, however quick, may be left on the sidelines. It also implies that when one has access to an AI algorithm and a flow of data, advantages accumulate over time and can’t be easily surmounted. For executives, entrepreneurs, policy makers, and everyone else, the best (and worst) about AI is yet to come.

要点

TAKEAWAYS

数据网络效应使人工智能变得更智能、更强大，并随着时间的推移不断完善和提高其准确性。人工智能可以利用客户互动、预测和反馈的力量，从每个用户体验中收集的数据积累中获益。

Data network effects have allowed AI to become smarter and more powerful, refining and improving its accuracy over time. AI can gain from an accumulation of data collected through each user’s experience by utilizing the power of customer interactions, predictions, and feedback.

✓ 反馈对于生成式人工智能算法的执行至关重要。如果没有持续的客户交互流，即使是最好的算法也无法长期保持智能。

✓  Feedback is crucial for generative AI algorithms to perform. Without constant streams of customer interactions, even the best algorithm won’t remain smart for long.

✓ 公司应定期进行细致的信息收集，以最大限度地发挥数据网络效应的优势。

✓  Companies should routinize meticulous gathering of information to maximize the benefits of data network effects.

✓ 每个人都应该考虑他们共享的数据。事实和反馈对于建立更好的预测至关重要，但您的数据的价值可能会被其他人获取。

✓  Everyone should consider the data they share. Facts and feedback are essential for building better predictions, but the value of your data can be captured by someone else.

改编自 hbr.org 上发布的内容，2023 年 3 月 14 日（产品#H07JCQ）。

Adapted from content posted on hbr.org, March 14, 2023 (product #H07JCQ).

风险与需求

Risk and Demand

关于风险，产生和消除不实和不准确的可能性有多大，破坏性有多大。受精？按需，除了当前的喧嚣之外，对这种输出的真正和可持续的需求是什么？

On risk, how likely and how damaging is the possibility of untruths and inaccuracies being generated and disseminated? On demand, what is the real and sustainable need for this kind of output, beyond the current buzz?

一起考虑这些变量很有用。在 2 × 2 矩阵中考虑它们可以对可能发生的情况提供更细致、一刀切的分析。事实上，不同行业和业务活动的风险和需求有所不同。我们在图 3-1中放置了一些常见的跨行业用例。

It’s useful to consider these variables together. Thinking of them in a 2 × 2 matrix provides a more nuanced, one-size-doesn’t-fit-all analysis of what may be coming. Indeed, risks and demands differ across different industries and business activities. We have placed some common cross-industry use cases in figure 3-1.

考虑一下您的业务职能或行业可能位于何处。对于您的用例，通过引入人工验证步骤可以降低多少风险？这会在多大程度上减缓进程并减少需求？

Think about where your business function or industry might sit. For your use case, how much is the risk reduced by introducing a step for human validation? How much might that slow down the process and reduce the demand?

左上角的方框——错误的后果相对较低，市场需求很高——将不可避免地发展得更快、更远。对于这些用例，公司有现成的动力去寻找解决方案，并且成功的障碍也较少。我们应该期望看到该技术的原始、即时利用以及第三方工具的结合，这些工具在其特定领域利用生成式人工智能及其 API。

The top-left box—where the consequence of errors is relatively low and market demand is high—will inevitably develop faster and further. For these use cases, there is a ready-made incentive for companies to find solutions, and there are fewer hurdles for their success. We should expect to see a combination of raw, immediate utilization of the technology as well as third-party tools that leverage generative AI and its APIs for their particular domain.

这种情况已经在营销领域发生，一些初创公司已经找到了应用法学硕士的创新方法来生成内容营销文案和创意，并取得了独角兽的地位。营销需要大量的创意产生和迭代、针对特定受众量身定制的消息传递以及生成丰富的文本消息可以吸引并影响观众。换句话说，有明确的用途和明确的需求。重要的是，还有大量的例子可以用来指导人工智能匹配风格和内容。另一方面，大多数营销文案并不注重事实，重要的事实可以在编辑时纠正。

This is happening already in marketing, where several startups have found innovative ways to apply LLMs to generate content marketing copy and ideas and have achieved unicorn status. Marketing requires a lot of idea generation and iteration, messaging tailored to specific audiences, and the production of text-rich messages that can engage and influence audiences. In other words, there are clear uses and demonstrated demand. Importantly, there’s also a wealth of examples that can be used to guide an AI to match style and content. On the other hand, most marketing copy isn’t fact-heavy, and the facts that are important can be corrected in editing.

图3-1

FIGURE 3-1

选择一个生成式人工智能项目

Picking a generative AI project

当您的公司决定从哪里开始探索生成式人工智能时，平衡风险和需求非常重要。思考这个问题的一种方法是问两个问题：“如果不实和不准确的信息产生并传播，会造成多大的损害？” （风险）以及“除了当前的喧嚣之外，对这种产出的真正和可持续的需求是什么？” （要求）。考虑使用此矩阵（包含常见的跨行业用例）来确定对您的公司最有价值、风险最小的应用程序。

As your company decides where to start exploring generative AI, it’s important to balance risk and demand. One way to think about that is to ask two questions: “How damaging would it be if untruths and inaccuracies were generated and disseminated?” (risk) and “What is the real and sustainable need for this kind of output, beyond the current buzz?” (demand). Consider using this matrix—populated with common, cross-industry use cases—to identify the most valuable, least-risky applications for your company.

查看矩阵，您可以发现还有其他机会受到较少的关注，例如学习。与营销一样，创建学习内容（出于我们的目的，让我们以内部企业学习工具为例）需要引人入胜且有效的文本，并清楚地了解受众的兴趣。还有可能可用于指导生成人工智能工具的内容。在现有文档的基础上，您可以要求它重写、综合和更新您必须的材料，以便更好地与不同的受众交流或使学习材料更适合不同的环境。

Looking at the matrix, you can find that there are other opportunities that have received less attention, for instance, learning. Like marketing, creating content for learning—for our purposes, let’s use the example of internal corporate learning tools—requires engaging and effective text and a clear understanding of its audience’s interests. There’s also likely content that can be used to guide a generative AI tool. Priming it with existing documentation, you can ask it to rewrite, synthesize, and update the materials you have to better speak to different audiences or to make learning material more adaptable to different contexts.

生成式人工智能的功能还可以让学习材料以不同的方式交付——融入日常工作流程或取代笨重的常见问题解答、庞大的知识中心和票务系统。

Generative AI’s capabilities could also allow learning materials to be delivered differently—woven into the flow of everyday work or replacing clunky FAQs, bulging knowledge centers, and ticketing systems.

上面高需求/低风险框中的其他用途遵循类似的逻辑：它们用于人们经常参与的任务，并且人工智能对事实的快速和松散的风险很低。以要求人工智能审阅文本为例：你可以给它提供草稿，给它一些指示（你想要更详细的版本、更柔和的语气、五点总结或如何使文本更简洁的建议），并查看其建议。作为第二双眼睛，该技术现在就可以使用了。如果你想要激发头脑风暴的想法——雇用现代多媒体设计师时要采取的步骤，或者给喜欢火车的 4 岁孩子买什么作为生日礼物——生成式人工智能将是一个快速、可靠和安全的选择，因为想法可能不会出现在最终产品中。

The other uses in the high-demand/low-risk box above follow similar logic: They’re for tasks where people are often involved, and the risk of AI playing fast and loose with facts are low. Take the examples of asking the AI to review text: You can feed it a draft, give it some instructions (you want a more detailed version, a softer tone, a five-point summary, or suggestions of how to make the text more concise), and review its suggestions. As a second pair of eyes, the technology is ready to use right now. If you want ideas to feed a brainstorm—steps to take when hiring a modern multimedia designer or what to buy a 4-year-old who likes trains for her birthday—generative AI will be a quick, reliable, and safe bet, as those ideas are likely not in the final product.

在矩阵中填写属于公司或团队工作一部分的任务可以帮助得出类似的相似之处。评估风险和需求并考虑特定任务的共同要素可以为您提供有用的起点，并帮助您建立联系并发现机会。它还可以帮助您了解哪些地方没有必要投入时间和资源。

Filling in the matrix with tasks that are part of your company’s or team’s work can help draw similar parallels. Assessing risk and demand and considering the shared elements of particular tasks can give you a useful starting point and help you draw connections and see opportunities. It can also help you see where it doesn’t make sense to invest time and resources.

其他三个象限不是你应该急于寻找生成式人工智能工具用途的地方。当需求较低时，人们就没有动力使用或开发该技术。今天，以莎士比亚海盗风格制作俳句可能会让我们开怀大笑，但这样的派对技巧不会让我们的注意力持续太久。在有需求但风险高的情况下，普遍的恐惧和监管将减缓进展的速度。考虑到您自己的 2 × 2 矩阵，您可以暂时将列出的用途放在一边。

The other three quadrants aren’t places where you should rush to find uses for generative AI tools. When demand is low, there’s little motivation for people to utilize or develop the technology. Producing haikus in the style of a Shakespearian pirate may make us laugh and drop our jaws today, but such party tricks will not keep our attention for very much longer. And in cases where there is demand but high risk, general trepidation and regulation will slow the pace of progress. Considering your own 2 × 2 matrix, you can put the uses listed there aside for the time being.

低风险仍然是风险

Low Risk Is Still Risk

温和的警告：即使在企业学习中，正如我们所说，风险很低，风险仍然存在。就像人类一样，生成式人工智能也容易受到偏见和错误的影响。如果你认为生成式人工智能系统的输出很好，并立即将它们分发给你的全体员工，那么就会存在很大的风险。您在速度和质量之间取得适当平衡的能力将受到考验。

A mild cautionary note: Even in corporate learning where, as we have argued, the risk is low, there is risk. Generative AI is vulnerable to bias and errors, just as humans are. If you assume the outputs of a generative AI system are good to go and immediately distribute them to your entire workforce, there is plenty of risk. Your ability to strike the right balance between speed and quality will be tested.

因此将初始输出作为第一次迭代。通过一两个更详细的提示来改进它。然后自己调整输出，添加现实世界的知识、细微差别，甚至艺术性和幽默感，而在一段时间内，只有人类才拥有这些。

So take the initial output as a first iteration. Improve on it with a more detailed prompt or two. And then tweak that output yourself, adding the real-world knowledge, nuance, even artistry and humor that, for a little while longer, only a human has.

要点

TAKEAWAYS

生成式人工智能几乎能够立即生成初稿并产生想法，但它也可能面临准确性和道德问题。企业在追求回报的过程中应如何应对风险？

Generative AI is able to produce first drafts and generate ideas virtually instantaneously, but it can also struggle with accuracy and ethical problems. How should companies navigate the risks in their pursuit of its rewards?

✓ 在选择用例时，公司需要平衡风险（产生和传播不实信息和不准确信息的可能性有多大，破坏性有多大？）和需求（除了当前的喧嚣之外，对这种输出的真正和可持续的需求是什么？ ？）。

✓  In picking use cases, companies need to balance risk (How likely and how damaging is the possibility of untruths and inaccuracies being generated and disseminated?) and demand (What is the real and sustainable need for this kind of output, beyond the current buzz?).

✓ 绘制风险和需求的 2 × 2 矩阵可以帮助企业选择最佳的生成式 AI 项目并提高成功机会。

✓  A 2 × 2 matrix that plots risk and demand can help companies choose the best generative AI projects and improve their chances of success.

✓ 公司应该进行适合矩阵高需求/低风险框的实验。其他三个象限不是公司应该急于寻找生成式人工智能工具用途的地方。

✓  Companies should run experiments that fit into the high-demand/low-risk box of the matrix. The other three quadrants aren’t places where companies should rush to find uses for generative AI tools.

笔记

NOTE

1. 1 . Ethan Mollick，“ChatGPT 是人工智能的引爆点”，hbr.org，2022 年 12 月 14 日， https:// hbr .org /2022 /12 /chatgpt -is -a -tipping -point -for -ai。

2. 1. Ethan Mollick, “ChatGPT Is a Tipping Point for AI,” hbr.org, December 14, 2022, https://hbr.org/2022/12/chatgpt-is-a-tipping-point-for-ai.

改编自 hbr.org 上发布的内容，2023 年 3 月 29 日（产品#H07J5S）。

Adapted from content posted on hbr.org, March 29, 2023 (product #H07J5S).

三种可能的未来

Three Possible Futures

如何为生成式人工智能做好准备

How to Prepare for Generative AI

创造力始终是任何公司创新过程及其竞争力的关键先决条件。不久前，创造力还是人类独有的事业。然而，正如我们所阐述的，生成式人工智能的到来即将改变这一切。为了做好准备，我们需要了解随之而来的威胁和挑战。一旦我们了解了要改变什么以及如何改变，我们就可以为未来做好准备，在未来，创意业务将成为人机的功能合作。下面，我们提供了三项建议，工作人员在采用生成式人工智能在当今的创意产业中创造商业价值和利润时应考虑这些建议。

Creativity has always been a critical prerequisite for any company’s innovation process and hence competitiveness. Not too long ago, the business of creativity was a uniquely human endeavor. However, as we’ve illustrated, the arrival of generative AI is about to change all this. To be prepared, we need to understand the accompanying threats and challenges. Once we understand what is to change and how, we can prepare for a future where the creativity business will be a function of human–machine collaborations. Below, we provide three recommendations that workers should consider as they adopt generative AI to create business value and profit in today’s creative industries.

笔记

NOTES

1. 1 . Nat Friedman，“GitHub Copilot 简介：您的 AI 配对程序员”，GitHub 博客，2021 年 6 月 29 日， https:// github .blog /2021 -06 -29 -introducing -github -copilot -ai -pair -programmer /。

2. 1. Nat Friedman, “Introducing GitHub Copilot: Your AI Pair Programmer,” GitHub blog, June 29, 2021, https://github.blog/2021-06-29-introducing-github-copilot-ai-pair-programmer/.

3. 2 . James Vincent，“BuzzFeed 表示将使用 OpenAI 的人工智能工具来个性化其内容”，The Verge，2023 年 1 月 21 日， https:// www .theverge .com /2023 /1 /26 /23572834 /buzzfeed -using -ai -工具-个性化-生成-内容-openai。

4. 2. James Vincent, “BuzzFeed Says It Will Use AI Tools from OpenAI to Personalize Its Content,” The Verge, January 21, 2023, https://www.theverge.com/2023/1/26/23572834/buzzfeed-using-ai-tools-personalize-generate-content-openai.

改编自 hbr.org 上发布的内容，2023 年 4 月 13 日（产品#H07LIA）。

Adapted from content posted on hbr.org, April 13, 2023 (product #H07LIA).

促进发散思维

Promote Divergent Thinking

生成式人工智能可以通过在远程概念之间建立关联并产生从中汲取的想法来支持发散性思维。下面是我们如何使用 Midjourney（一种文本到图像算法，可以检测图像之间的类比相似性）根据人类的文本提示生成新颖的产品设计的示例。 （我们在本文中使用了 Midjourney、ChatGPT 和 Stable Diffusion 作为示例，但它们只是现在可用的众多生成式 AI 工具中的几个。）我们要求 Midjourney 创建一个图像，将大象和蝴蝶结合在一起，产生了我们称之为“幻蝇”的嵌合体。

Generative AI can support divergent thinking by making associations among remote concepts and producing ideas drawn from them. Here’s an example of how we used Midjourney, a text-to-image algorithm that can detect analogical resemblances between images, to generate novel product designs based on textual prompts from a human. (We used Midjourney, ChatGPT, and Stable Diffusion for the examples in this article, but they are just a few of a host of generative AI tools that are now available.) We asked Midjourney to create an image that combined an elephant and a butterfly, and it produced the chimera we dubbed “phantafly.”

然后，我们使用 Midjourney 中的详细渲染来激发稳定扩散（另一种流行的文本到图像模型）中的提示。 Stable Diffusion 为不同的产品类别产生了一系列创意，包括椅子和手工巧克力糖果（见图5-1和5-2）。

We then used the detailed rendering from Midjourney to inspire prompts in Stable Diffusion, another popular text-to-image model. Stable Diffusion generated a range of ideas for different product categories, including chairs and artisanal chocolate candies (see figures 5-1 and 5-2).

以这种方式快速且廉价地生产大量设计使公司能够评估广泛的设计快速了解一系列产品概念。例如，一家服装公司使用生成式人工智能来创造新的 T 恤设计，可以紧跟潮流，并为客户提供不断变化的产品选择。

Rapidly and inexpensively producing a plethora of designs in this way allows a company to evaluate a wide range of product concepts quickly. For example, a clothing company that uses generative AI to create new designs for T-shirts could stay on top of trends and offer a constantly changing selection of products to customers.

图 5-1

FIGURE 5-1

Stable Diffusion 的 Phantafly 风格椅子概念

Phantafly-inspired chair concepts by Stable Diffusion

图5-2

FIGURE 5-2

Stable Diffusion 以 Phantafly 为灵感的手工巧克力概念

Phantafly-inspired artisanal chocolate concepts by Stable Diffusion

考虑另一个例子，说明这项技术如何将想法联系起来，创造出个人或团队可能永远不会想出的概念。我们使用了 ChatGPT，一种生成式人工智能，称为一个大的语言模型，来指导想法的产生。我们要求它通过三联过程产生想法——连接三个不同的实体（双联创造力技术的延伸）。我们的团队给了 ChatGPT 以下提示：“你将扮演一个创意者的角色。您将随机生成 10 个常见名词。然后，您将随机选择 10 个名词中的任意两个。然后你会问我第三个名词。你将通过组合或关联你确定的两个名词和我确定的名词来产生一个商业想法。”

Consider another example of how this technology can connect ideas to create concepts that an individual or a team might never have come up with themselves. We used ChatGPT, a type of generative AI known as a large language model, to guide the production of ideas. We asked it to generate ideas through a process of trisociation—connecting three distinct entities (an extension of the bisociation creativity technique). Our team gave ChatGPT the following prompt: “You will play the role of an ideator. You will randomly generate 10 common nouns. You will then randomly select any two of the 10 nouns. You will then ask me for a third noun. You will generate a business idea by combining or associating the two nouns you identified and the noun I identified.”

ChatGPT 生成了名词food和technology。当出现提示时，我们提供了额外的名词car。 ChatGPT 在短时间内提出了以下商业构想：“一种智能送餐服务，使用自动驾驶汽车将餐食运送给顾客。技术方面可能涉及使用人工智能来优化配送路线、实时跟踪食品温度以及为客户提供订单状态的实时更新。该服务可以针对忙碌的专业人士和家庭，他们希望在不牺牲口味和质量的情况下选择方便、健康的膳食。”

ChatGPT generated the nouns food and technology. When prompted, we provided the additional noun car. ChatGPT produced the following business idea in short order: “A smart food-delivery service that uses self-driving cars to transport meals to customers. The technology aspect could involve using AI to optimize delivery routes, track food temperature in real time, and provide customers with real-time updates on the status of their orders. The service could target busy professionals and families who want convenient and healthy meal options without sacrificing taste and quality.”

在另一轮中，ChatGPT 生成了名词Airlines和chair。当提示时，我们提供了University，ChatGPT 提出了一个商业理念，为学生提供一种便捷、经济高效的方式和学者可以前往世界各地参加会议和研讨会，并在飞行期间访问教育书籍图书馆。它提议将公司命名为 Fly and Study 或 Edu-Fly。

In a separate round, ChatGPT produced the nouns airline and chair. When prompted, we provided university, and ChatGPT came up with a business concept that provides a convenient, cost-effective way for students and academics to travel to conferences and workshops around the world along with access to a library of educational books during the flight. It proposed that the company be called Fly and Study or Edu-Fly.

挑战专业偏见

Challenge Expertise Bias

在新产品开发的早期阶段，生成式人工智能创建的非典型设计可以激发设计师超越他们对产品在形式和功能方面的可能性或期望的先入之见。这种方法可以带来人类使用传统方法可能从未想象过的解决方案，在传统方法中，首先确定功能，然后设计形式来适应它们。这些输入可以帮助克服偏见，例如设计固定（过度依赖标准设计形式）、功能固定（缺乏想象传统用途之外的用途的能力）以及个体效应（个人以前的经验阻碍他们考虑）解决问题的新方法。

During the early stages of new-product development, atypical designs created by generative AI can inspire designers to think beyond their preconceptions of what is possible or desirable in a product in terms of both form and function. This approach can lead to solutions that humans might never have imagined using a traditional approach, where the functions are determined first and the form is then designed to accommodate them. These inputs can help overcome biases such as design fixation (an overreliance on standard design forms), functional fixedness (a lack of ability to imagine a use beyond the traditional one), and the Einstellung effect, where individuals’ previous experiences impede them from considering new ways to solve problems.

这是此过程的一个示例。我们要求 Stable Diffusion 生成受螃蟹启发的玩具的通用设计，但没有提供任何功能规格。然后我们在看到之后想象了功能能力设计。例如，图5-3所示的螃蟹玩具系列中，左上角的图像可以开发成爬墙玩具；旁边的图像可能是一个可以发射小球穿过房间的玩具。靠近中心的盘子里的螃蟹可以成为宠物的慢食盘。

Here’s an example of this process. We asked Stable Diffusion to generate generic designs of crab-inspired toys but provided it with no functional specifications. Then we imagined functional capabilities after seeing the designs. For instance, in the collection of crab-inspired toys shown in figure 5-3, the image in the top left could be developed into a wall-climbing toy; the image next to it could be a toy that launches a small ball across a room. The crab on a plate near the center could become a slow-feeder dish for pets.

这并不是一种完全新颖的方式来推出不寻常的产品：大部分的建筑和骑行迪士尼世界等主题公园的功能是由再现故事中的场景和人物的愿望驱动的。但生成式人工智能工具可以帮助启动公司的富有想象力的设计。

This is not a completely novel way to come up with unusual products: Much of the architecture and ride functionality in theme parks such as Disney World has been driven by a desire to recreate scenes and characters from a story. But generative AI tools can help jump-start a company’s imaginative designs.

图 5-3

FIGURE 5-3

Stable Diffusion 以螃蟹为灵感的玩具概念

Crab-inspired toy concepts by Stable Diffusion

协助创意评估

Assist in Idea Evaluation

生成式人工智能工具可以在创新前端的其他方面提供帮助，包括提高想法的特异性、评估想法以及有时将它们组合起来。考虑一个创新挑战，其目标是找出尽量减少食物浪费的方法。 ChatGPT 评估了三个原始想法的优缺点：(1) 具有动态有效期的包装（标签根据储存地点的环境条件自动更改日期或颜色）； (2) 帮助用户捐赠食物的应用程序； (3) 开展宣传活动，教育人们了解保质期的类型以及它们在新鲜度和适用性方面所代表的意义。 ChatGPT 对利弊进行了平衡的分析，这反映了我们对两个感兴趣的人之间讨论这些想法的优点的交流所期望的结果。

Generative AI tools can assist in other aspects of the front end of innovation, including by increasing the specificity of ideas and by evaluating ideas and sometimes combining them. Consider an innovation challenge where the goal is to identify ways to minimize food waste. ChatGPT assessed the pros and cons of three raw ideas: (1) packaging with dynamic expiration dates (labels that automatically change either the dates or colors based on the environmental conditions in the places where they are stored); (2) an app to help users donate food; and (3) a campaign to educate people on types of expiration dates and what they represent in terms of freshness and fitness for use. ChatGPT produced a balanced analysis of the pros and cons that mirrored what we might expect from an exchange between two interested persons discussing the merits of such ideas.

例如，当 ChatGPT 评估动态过期日期打包的概念时，它确定它将帮助消费者更好地了解产品的保质期，并鼓励食品制造商生产小批量的产品，以便在杂货店货架上更频繁地补充。此外，ChatGPT 指出，动态有效期可能需要对制造和包装过程进行重大改变，因此可能会增加制造商和消费者的成本。

When ChatGPT evaluated the concept of dynamic expiration-date packaging, for instance, it determined that it would help consumers better understand the shelf life of products and encourage food manufacturers to produce smaller batches that would be replenished more frequently on grocery shelves. In addition, ChatGPT pointed out that dynamic expiration dates might require significant changes to the manufacturing and packaging process and as a result, could increase the costs to both manufacturers and consumers.

ChatGPT 确定，这款食物捐赠应用程序可以鼓励人们在食物变质之前用完，并通过向有需要的人提供未开封的可食用食物来减少食物浪费。它警告说，该应用程序可能需要庞大的用户群才能发挥作用，而且来自各种不受监管的来源的食品运输和分配可能会带来安全问题。

ChatGPT determined that the food-donation app could encourage people to use up their food before it goes bad and reduce food waste by giving unopened, edible food to those in need. It cautioned that the app could require a large user base to be effective and that the transportation and distribution of food from a wide variety of unregulated sources could pose safety concerns.

它表示，针对消费者的教育计划的优点是提高消费者对不同过期标签含义的认识，并帮助他们就食品购买和浪费做出更明智的决定。但 ChatGPT 警告称，该教育计划可能过于复杂，因为所有食品的保质期并未标准化。它还警告说，如果该计划范围广泛，特别是当它涉及广泛的活动或教育材料时，对用户进行不同类型的到期日期的教育可能会代价高昂。

It stated that the pros of an education program for consumers were increasing consumer awareness of the meaning of different expiration labels and helping them make more-informed decisions about food purchases and waste. But ChatGPT warned that this education program could be overly complex because expiration dates are not standardized across all food products. And it cautioned that educating users on different types of expiration dates can be costly if the program is broad in scope, particularly if it involves widespread campaigns or educational materials.

生成式人工智能可以超越简单的利弊，帮助人类评估创造力的维度，例如新颖性、可行性、特异性、影响力和可操作性。我们要求 ChatGPT 使用这些标准评估相同的示例。

Generative AI can go beyond simple pros and cons and help humans evaluate dimensions of creativity such as novelty, feasibility, specificity, impact, and workability. We asked ChatGPT to assess the same examples using these criteria.

以下是其对动态有效期包装概念的评估：

Here is its assessment of the dynamic expiration-date packaging concept:

• 新颖性：有点新颖。人们一直在努力开发可以延长食品保质期的智能包装，但动态有效期的想法是一个新兴概念。
• Novelty: Somewhat novel. There have been efforts to develop smart packaging that can extend the shelf life of food products, but the idea of dynamic expiration dates is an emerging concept.
• 可行性：实施起来可能具有挑战性，因为它需要开发新的包装材料和技术。它还需要与食品制造商、监管机构和分销商合作。
• Feasibility: May be challenging to implement as it requires the development of new packaging materials and technologies. It would also require collaboration with food manufacturers, regulators, and distributors.
• 特殊性：它的特殊性在于它通过为消费者提供更准确的食品有效期来解决食物浪费问题。
• Specificity: It’s specific in that it addresses the problem of food waste by providing consumers with more-accurate expiration dates for their food.
• 影响：可能对减少食物浪费和提高食品安全产生重大影响。如果提供更准确的保质期，消费者将更有可能在食物变质之前使用它们。
• Impact: Could have a significant impact on reducing food waste and improving food safety. When provided with more-accurate expiration dates, consumers would be more likely to use their food before it goes bad.
• 可行性：它很可能需要大量资源和专业知识来开发和实施，并且可能需要几年时间才能充分发挥其潜力。
• Workability: It would most likely require significant resources and expertise to develop and implement and may take several years to fully realize its potential.

以下是 ChatGPT 对捐赠即将过期食物的应用程序的评估：

Here is ChatGPT’s assessment of the app for donating food close to expiration:

• 新颖性：不是特别新颖。已经有许多项目和组织致力于减少食物浪费并将剩余食物捐赠给有需要的人。
• Novelty: Not particularly novel. There are already many programs and organizations dedicated to reducing food waste and donating surplus food to those in need.
• 可行性：高度可行。有许多已建立的模式和基础设施来支持食品捐赠工作。
• Feasibility: Highly feasible. There are many established models and an infrastructure in place to support food donation efforts.
• 特异性：减少食物浪费和解决粮食不安全问题的具体目标。
• Specificity: Specific in its aim to reduce food waste and address issues of food insecurity.
• 影响：有潜力产生重大影响，因为它可以解决食物浪费和饥饿问题。通过将剩余粮食重新分配给有需要的人，可以帮助减少粮食浪费，同时改善弱势群体的粮食安全。
• Impact: Has the potential to make a significant impact as it addresses both food waste and hunger. By redirecting surplus food to those in need, it can help reduce food waste while also improving food security for vulnerable populations.
• 可操作性：可操作性强，可以快速且相对较低的成本实施。这需要与食品制造商建立合作伙伴关系读者、经销商、食品银行以及收集和分发捐赠食品的志愿者。
• Workability: Highly workable in that it can be implemented quickly and at a relatively low cost. It would require partnerships with food manufacturers, distributors, and food banks, as well as volunteers to collect and distribute the donated food.

最后，以下是 ChatGPT 如何评估教育用户了解不同类型的到期日期的想法：

Finally, here is how ChatGPT evaluated the idea to educate users on different types of expiration dates:

• 新颖性：不是特别新颖。人们一直在努力教育消费者了解保质期的含义以及如何正确储存食品以延长其保质期。
• Novelty: Not particularly novel. There have been efforts to educate consumers on the meaning of expiration dates and how to properly store food to extend its shelf life.
• 可行性：高度可行。它可以通过多种方式实施，例如公众意识活动、教育材料和媒体宣传。
• Feasibility: Highly feasible. It can be implemented through various means, such as public awareness campaigns, educational materials, and media outreach.
• 特殊性：其目的是教育消费者不同的保质期以及如何正确储存食物以减少食物浪费。
• Specificity: Specific in its aim to educate consumers on the different expiration dates and how to properly store food to reduce food waste.
• 影响：潜在重大。教育消费者了解不同的保质期以及如何正确储存食物将使他们更加注意食物浪费并采取措施减少食物浪费。
• Impact: Potentially significant. Educating consumers on the different expiration dates and how to properly store food will make them more mindful of food waste and take steps to reduce it.
• 可操作性：可操作性强，可以以相对较低的成本实施。这需要与食品工业团体、政府合作开发和分发教育材料的机构和其他组织。
• Workability: Highly workable and can be implemented at a relatively low cost. It would require collaboration with food industry groups, government agencies, and other organizations to develop and distribute educational materials.

使用 ChatGPT 的评估，可以相对容易地评估这三个概念，或者根据最重要的标准快速评分或组织它们。

Using ChatGPT’s assessments, it would be relatively easy to evaluate these three concepts, or quickly score or organize them based on the criteria that matter most.

支持想法细化

Support Idea Refinement

生成式人工智能工具可以解决创意竞赛中面临的一个重要挑战：组合或合并大量创意以产生更强大的创意。例如，试图赢得合同或拨款来开发减少食物浪费计划的组织可以使用这些工具来结合前面讨论的三个概念。我们分两步完成此操作。首先，我们要求 ChatGPT 充实每个概念：

Generative AI tools can solve an important challenge faced in idea contests: combining or merging a large number of ideas to produce much stronger ones. For example, an organization that is trying to win a contract or a grant to develop a food-waste-reduction program can use the tools to combine the three concepts discussed earlier. We did this in two steps. First, we asked ChatGPT to flesh out each of the concepts:

• “动态保质期是一种旨在减少食物浪费的解决方案。这涉及将技术融入包装中，以实时跟踪和调整产品的保质期。通过使用传感器，可以根据环境因素的变化更新有效期，从而使消费者能够更准确地估计产品何时会变质。”
• “Dynamic expiration dates are a solution aimed at reducing food waste. This involves incorporating technology into packaging that can track and adjust the shelf life of products in real time. By using sensors, the expiration date can be updated based on changes in environmental factors, providing consumers with a more accurate estimate of when a product will go bad.”
• “一款让消费者能够捐赠接近保质期食物的应用程序旨在通过向有需要的人提供未开封的可食用食物来减少食物浪费。该应用程序将拥有剩余食物的个人与可以使用这些食物的当地组织联系起来。通过利用技术促进食物捐赠，该应用程序可以帮助减少食物浪费并为有需要的人提供帮助。”
• “An app that enables consumers to donate food close to its expiration date aims to reduce food waste by giving unopened, edible food to those in need. The app connects individuals who have surplus food to local organizations that can use it. By using technology to facilitate food donations, the app can help reduce food waste and provide for those in need.”
• “对用户进行不同类型的保质期教育可以帮助用户在购买和消费食品、减少浪费和省钱方面做出明智的决定。通过澄清“此日期前最佳”、“此日期前使用”和其他过期标签的含义，教育活动可以帮助用户更好地了解如何确定食物是否可以安全食用以及何时需要处理掉。”
• “Educating users on the different types of expiration dates can help users make informed decisions about purchasing and consuming food, reducing waste, and saving money. By clarifying the meaning of ‘best before,’ ‘use by,’ and other expiration labels, an education campaign can help users better understand how to determine if food is safe to eat and when it is time to dispose of it.”

接下来，我们要求 ChatGPT 将所有三个想法合并到一个食物垃圾计划中。它简短地描述了以下内容：“全面的减少食物浪费的解决方案包括创建具有动态有效期的智能包装、建立捐赠剩余食物的计划以及教育公众了解食物有效期的重要性。智能包装帮助消费者就食品新鲜度做出明智的决定，捐赠计划确保剩余食物得到充分利用，教育提高人们对食物浪费影响的认识。该解决方案有潜力通过赋予消费者权力、与行业和政府合作以及更好地利用剩余食物来显着减少食物浪费并创造更可持续的未来。”

Next, we asked ChatGPT to merge all three ideas into one food-waste program. In short order it produced the following description: “The comprehensive food-waste-reduction solution involves creating smart packaging with dynamic expiration dates, establishing a program for donating surplus food, and educating the public on the importance of understanding food expiration dates. The smart packaging helps consumers make informed decisions about food freshness, the donation program ensures surplus food is put to good use, and education raises awareness about the impact of food waste. This solution has the potential to significantly reduce food waste and create a more sustainable future by empowering consumers, working with industry and government, and making better use of surplus food.”

促进与用户之间的协作

Facilitate Collaboration with and Among Users

在开发新产品创意或设计时，生成式人工智能可以促进公司设计师和预期产品用户之间以及用户之间的协作。换句话说，该技术使新产品的共同创造变得更加容易且成本更低。例如，企业可以为用户提供生成设计的工具，然后根据客户的喜好创建个性化的产品版本。或者，用户的设计可以发布在众包平台上，由社区对其进行评估。然后，其他合作者可以进一步开发顶级设计。

When developing new product ideas or designs, generative AI can facilitate collaborations between a company’s designers and users of a prospective product and among users themselves. In other words, the technology makes co-creation of new offerings much easier and less expensive. For example, a business can give users access to tools to generate designs and then create a personalized version of the product based on the customer’s preferences. Alternatively, users’ designs could be posted on a crowdsourcing platform where they are evaluated by the community. The top designs could then be further developed by additional collaborators.

为了说明其潜力，我们展示了如何设计飞行汽车（人们已经尝试开发这种汽车 100 多年，但没有取得多大成功）。我们给稳定扩散这个提示：“设计一款既可以飞行又可以在路上行驶的产品，即飞行汽车。”稳定扩散产生了几种设计，我们选择了我们认为最有前途的一种：图 5-4右下角的车辆。

To illustrate the potential, we show how a flying car—something people have been trying to develop for more than 100 years without much success—might be designed. We gave Stable Diffusion this prompt: “Design a product that can fly but also drive on the road, a flying automobile.” Stable Diffusion generated several designs, and we selected what we considered to be the most promising one: the vehicle in the lower right corner of figure 5-4.

然后我们要求 Stable Diffusion 采用该设计并重新构想这个概念，使汽车“像一只机器鹰”。图 5-5显示了生成式人工智能程序很快就产生了——从左上角最像机器鹰的设计到右下角更可行的飞行汽车概念。

Then we asked Stable Diffusion to take that design and reimagine the concept so that the car “resembles a robot eagle.” Figure 5-5 shows the variations that the generative AI program quickly produced—from the top left design that looks most like a robot eagle to the more feasible concept of a flying automobile in the lower right corner.

图 5-4

FIGURE 5-4

Stable Diffusion 的飞行汽车概念

Stable Diffusion’s concepts of a flying automobile

图5-5

FIGURE 5-5

Stable Diffusion 的类似机器鹰的飞行汽车概念

Stable Diffusion’s concepts of a flying automobile that resembles a robot eagle

第二个示例说明了设计人员如何使用此类工具就结构设计的主题变化进行协作。他们从人工智能生成的飞行汽车设计开始，并要求该工具生产类似蜻蜓、老虎、乌龟和鹰的版本（见图5-6）。

A second example illustrates how designers can use such tools to collaborate on thematic variations of a structural design. They began with a flying-automobile design generated by AI and asked the tool to produce versions that resembled a dragonfly, a tiger, a tortoise, and an eagle (see figure 5-6).

另一种方法是人类协作者使用 ChatGPT 等工具来开发产品的细节，然后使用 Stable Diffusion 等工具根据一系列相互构建的提示获得视觉设计。我们给 ChatGPT 的提示与我们给 Stable Diffusion 的提示类似：“描述一种既可以飞行又可以在路上行驶的产品，即飞行汽车。”

An alternative approach is for human collaborators to use a tool like ChatGPT to develop details of the product and then use one like Stable Diffusion to obtain visual designs based on a series of prompts that build on one another. We gave ChatGPT a similar prompt to what we had given to Stable Diffusion: “Describe a product that can fly but also drive on the road, a flying automobile.”

图5-6

FIGURE 5-6

人工智能生成的飞行汽车设计，形状类似于蜻蜓、老虎、乌龟和鹰

AI-generated designs of a flying car that resemble a dragonfly, a tiger, a tortoise, and an eagle

ChatGPT 提供了这样的描述：“飞行汽车是一款时尚且具有未来感的车辆，专为终极冒险而打造。它具有时尚的外观跑车拥有平滑的曲线和抛光的外观，但带有隐藏的转子，使其能够飞行。”当我们对稳定扩散进行描述时，它提供了如图 5-7所示的图像。

ChatGPT provided this description: “The flying automobile is a sleek and futuristic vehicle that is built for the ultimate adventure. It has the appearance of a stylish sports car with smooth curves and polished exterior but with hidden rotors that allow it to take flight.” When we gave that description to Stable Diffusion, it provided the image shown in figure 5-7.

图 5-7

FIGURE 5-7

使用稳定扩散从 ChatGPT 描述生成设计的飞行汽车设计

Flying automobile design using Stable Diffusion to generate a design from a ChatGPT description

接下来，我们要求 ChatGPT 重新构思描述，以包含产品必须类似于蜻蜓并具有夜间飞行的照明标记的信息。其回复如下：“凭借其细长的车身、伸展的机翼和隐藏的旋翼，该飞行器让人想起一只栩栩如生的蜻蜓。这沿着机翼和车身的发光标记创造了令人惊叹的视觉效果，有助于使车辆在黑暗中可见。”稳定扩散将这种描述翻译成各种版本，保持了可行的设计，并根据蜻蜓翅膀的图案添加了照明元素。图 5-8中的图像是示例。

Next we asked ChatGPT to reimagine the description to include the information that the product must resemble a dragonfly and have illumination markers for flying at night. It came back with the following: “With its slender body, extended wings, and hidden rotors, the vehicle is reminiscent of a dragonfly come to life. The illuminated markers located along the wings and body create a stunning visual effect, helping to make the vehicle visible in the darkness.” Stable Diffusion translated that description into various versions that maintained the feasible design and added elements of illumination based on the pattern of a dragonfly’s wings. The images in figure 5-8 are examples.

图5-8

FIGURE 5-8

融入蜻蜓细节和照明的设计变体

Variations on the design that incorporate dragonfly details and illumination

•••

•  •  •

人类拥有无限的创造力。然而，以书面或视觉形式传达概念的挑战限制了很多人贡献新想法。生成式人工智能可以消除这个障碍。与任何真正的创新能力一样，毫无疑问会遇到阻力。长期存在的创新流程必须改变。在旧的做事方式中拥有既得利益的人——尤其是那些担心自己会被淘汰的人——将会抵制。但其优势——有机会大幅增加组织内部和外部想法的数量和新颖性——将使这一旅程变得值得。生成式人工智能的最大潜力不是取代人类，而是取代人类。它旨在帮助人类通过个人和集体努力创造迄今为止难以想象的解决方案。它可以真正实现创新民主化。

Humans have boundless creativity. However, the challenge of communicating their concepts in written or visual form restricts vast numbers of people from contributing new ideas. Generative AI can remove this obstacle. As with any truly innovative capability, there will undoubtedly be resistance to it. Long-standing innovation processes will have to change. People with vested interests in the old way of doing things—especially those worried about being rendered obsolete—will resist. But the advantages—the opportunities to dramatically increase the number and novelty of ideas from both inside and outside the organization—will make the journey worthwhile. Generative AI’s greatest potential is not replacing humans; it is to assist humans in their individual and collective efforts to create hitherto unimaginable solutions. It can truly democratize innovation.

要点

TAKEAWAYS

生成式人工智能具有增强人类创造力的潜力。它使设计师能够从多个角度研究概念，发散性思考，超越他们的视野自己的假设，并使用数据驱动的见解来质疑这些假设。

Generative AI has the potential to augment human creativity. It enables designers to investigate concepts from several perspectives, think divergently, see beyond their own assumptions, and use data-driven insights to question those assumptions.

✓ 人工智能可以帮助解决与创造力相关的问题，例如评估过载、专业知识偏差、细节不足以及难以理解大局。

✓  AI can help solve creativity-related problems like assessment overload, expertise bias, insufficient details, and trouble understanding the bigger picture.

✓ 生成式人工智能可以通过评估新概念和现有未开发概念的组合来支持想法的检查和改进。

✓  Generative AI can support the examination and improvement of ideas by evaluating fresh concepts and combinations of already-existing undeveloped concepts.

✓ 这些技术鼓励用户参与新产品的共同开发。

✓  These technologies encourage user participation in the codevelopment of new products.

改编自 《哈佛商业评论》2023 年 7 月至 8 月的一篇文章（产品 #R2304C）。

Adapted from an article in Harvard Business Review, July–August 2023 (product #R2304C).

什么是可能的

What’s Possible

在解决如何做之前，请考虑生成式人工智能可以为销售组织做什么。

Before addressing the how, consider what generative AI can do for sales organizations.

价值之旅

The Journey to Value

生成式人工智能相对较新，并且发展迅速。缺乏确定其角色、培训的人才和微调模型，以及开发和实施应用程序。人们必须找到一种途径来防范虚假挑战，快速实现价值，并在控制成本的同时交付成果。

Generative AI is relatively new and evolving rapidly. There is a shortage of talent for defining its role, training and fine-tuning models, and developing and implementing applications. One must find pathways that guard against falsehood challenges, realize value quickly, and deliver results while keeping costs under control.

人工智能是生产力助手还是销售人员的替代品？

Is AI a Productivity Aid or a Substitute for Salespeople?

我们预计生成式人工智能将为几乎每个销售人员和销售经理提供数字助理。这些工具已经在帮助文案撰写者起草内容和计算机程序员编写代码，将他们的工作效率提高 50% 或更多。它对销售人员也能起到同样的作用。

We expect generative AI is to power digital assistants for nearly every salesperson and sales manager. These tools are already helping copywriters draft content and computer programmers write code, boosting their productivity by 50% or more. It can do the same for salespeople.

人工智能已经让客户自助服务变得更加强大，内部销售也变得更加强大。消费者是越来越多地使用数字技术自行研究产品和服务。

AI is already making customer self-service more powerful, and inside sales more potent. Consumers are increasingly using digital technology to research products and services on their own.

电子商务也在 B2B 领域蓬勃发展。即使在复杂的销售中，数字化也发挥着越来越重要的作用，承担着潜在客户开发和优先级排序、产品信息共享和配置以及订单下达等任务。数字化和内部销售不可避免地继续接管现场销售人员过去所做的许多任务，特别是对于熟悉的采购。

E-commerce has taken off in the B2B world too. Even in complex sales, digital plays an increasing role, taking on tasks such as lead generation and prioritization, product information sharing and configuring, and order placement. Inexorably, digital and inside sales continue to take over many tasks that field salespeople used to do, especially for familiar purchases.

然而，新的复杂产品仍然需要销售人员能够识别感知和潜在需求、定制解决方案并驾驭复杂的采购组织。是的，人工智能将夺走销售人员的任务，并在复杂的情况下进一步缩小他们的角色。与此同时，销售人工智能技术的公司将组建庞大的销售队伍，以抓住迫在眉睫的巨大而复杂的机会。

However, new and complex offerings still require salespeople who can identify perceived and latent needs, tailor solutions, and navigate complex buying organizations. Yes, AI will take tasks away from salespeople and narrow their role even more on complex situations. At the same time, the companies that sell AI technologies will create large sales forces to capture the looming massive and complex opportunities.

要点

TAKEAWAYS

生成式人工智能可以让销售代表和经理腾出时间专注于更多增值活动，从而实现销售转型。当生成式人工智能适当时集成到销售流程中，肯定会提高生产力。

Generative AI can transform sales by freeing up time for sales representatives and managers to focus on more value-adding activities. When generative AI is properly integrated into sales processes, it is certain to increase productivity.

✓ 这些工具可以帮助制定重要的客户策略、扭转管理蔓延、提供个性化内容和产品、回复提案请求电子邮件、与客户合作等等。

✓  These tools can assist in developing important account strategies, reversing administrative creep, providing personalized content and product offers, responding to proposal request emails, working with customers, and more.

✓ 在整合这些技术的过程中，销售人员和销售经理需要制定策略来处理不一致和不准确的问题，快速实现价值，并在控制成本的同时交付成果。

✓  On the pathway to integrating these technologies, salespeople and sales managers will need strategies for dealing with inconsistency and inaccuracy, realizing value rapidly, and delivering results while controlling costs.

✓ 人工智能正在迅速成为必要的数字助理，但对于复杂的产品，知识渊博的销售人员仍然是必要的。

✓  AI is quickly becoming a necessary digital assistant, but for complex products, knowledgeable salespeople are and will continue to be needed.

改编自 hbr.org 上发布的内容，2023 年 3 月 31 日（产品#H07JGX）。

Adapted from content posted on hbr.org, March 31, 2023 (product #H07JGX).

生成式人工智能如何适应当今的法律环境

Where Generative AI Fits into Today’s Legal Landscape

尽管生成式人工智能对于市场来说可能是新事物，但现有法律对其使用具有重大影响。法院正在整理如何应用成文法律。存在侵权和使用权问题、人工智能生成作品所有权的不确定性、训练数据中未经许可的内容以及用户是否应该能够通过直接引用其他创作者的版权和商标作品来提示这些工具的问题。未经他们允许的名字。

Though generative AI may be new to the market, existing laws have significant implications for its use. Courts are sorting out how the laws on the books should be applied. There are infringement and right-of-use issues, uncertainty about ownership of AI-generated works, and questions about unlicensed content in training data and whether users should be able to prompt these tools with direct reference to other creators’ copyrighted and trademarked works by name without their permission.

这些索赔已经被提起诉讼。在 2022 年底提起的一起案件中，Andersen 诉 Stability AI 等人。之后，三名艺术家组成集体起诉多个生成式 AI 平台，理由是 AI 在未经许可的情况下使用他们的原创作品，以他们的风格训练他们的 AI。因此，这些平台允许用户生成可能不足以对艺术家现有受保护作品进行改造的作品，因此将成为未经授权的衍生作品。如果法院发现人工智能的作品未经授权且具有衍生性，则可能会受到严厉的侵权处罚。

These claims are already being litigated. In a case filed in late 2022, Andersen v. Stability AI et al., three artists formed a class to sue multiple generative AI platforms on the grounds that the AI was using their original works without license to train their AI in their styles. The platforms were thus allowing users to generate works that might be insufficiently transformative from the artists’ existing protected works and, as a result, would be unauthorized derivative works. If a court finds that the AI’s works are unauthorized and derivative, substantial infringement penalties can apply.

2023 年提起的类似案件声称，公司使用包含数千甚至数百万未经许可作品的数据湖来训练人工智能工具。图像许可服务机构 Getty 对 Stable Diffusion 的创作者提起诉讼，指控其照片使用不当，侵犯了其带水印照片集的版权和商标权。

Similar cases filed in 2023 bring claims that companies trained AI tools using data lakes with thousands—or even many millions—of unlicensed works. Getty, an image licensing service, filed a lawsuit against the creators of Stable Diffusion alleging the improper use of its photos, violating both copyright and trademark rights it has in its watermarked photograph collection.

在每一个案件中，法律体系都被要求澄清知识产权法下“衍生作品”的范围，并且根据管辖权的不同，不同的联邦巡回法院可能会做出不同的解释。这些案件的结果预计将取决于对合理使用原则的解释，该原则允许未经所有者许可而使用受版权保护的作品“用于批评（包括讽刺）、评论、新闻报道、教学（包括多份副本）等目的用于课堂使用）、学术或研究”，以及以非预期方式对受版权保护的材料进行变革性使用。

In each of these cases, the legal system is being asked to clarify the bounds of what is a “derivative work” under intellectual property laws—and depending on the jurisdiction, different federal circuit courts may respond with different interpretations. The outcome of these cases is expected to hinge on the interpretation of the fair use doctrine, which allows copyrighted work to be used without the owner’s permission “for purposes such as criticism (including satire), comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research,” and for a transformative use of the copyrighted material in a manner for which it was not intended.

这并不是技术和版权法第一次发生冲突。谷歌成功地在一场诉讼中为自己辩护，认为变革性使用允许从书籍中抓取文本来创建其搜索引擎，目前，这一决定仍然是先例。

This isn’t the first time technology and copyright law have crashed into each other. Google successfully defended itself against a lawsuit by arguing that transformative use allowed for the scraping of text from books to create its search engine, and for the time being, this decision remains precedential.

但还有其他非技术案例可能会影响生成式人工智能产品的处理方式。 2023 年，美国最高法院向安迪·沃霍尔基金会提起诉讼，由摄影师林恩·戈德史密斯 (Lynn Goldsmith) 提起，该人获得了已故音乐家普林斯 (Prince) 的一张照片的许可，这可能会完善美国版权法，解决一件艺术品何时与其他艺术品有足够不同的问题。其源材料是否具有明确的“变革性”，以及法院在评估这种变革时是否可以考虑衍生作品的含义。法院裁定沃霍尔的作品不属于合理使用，这可能会给人工智能生成的作品带来麻烦。

But there are other, nontechnological cases that could shape how the products of generative AI are treated. A 2023 case before the U.S. Supreme Court against the Andy Warhol Foundation—brought by photographer Lynn Goldsmith, who had licensed an image of the late musician, Prince—may refine U.S. copyright law on the issue of when a piece of art is sufficiently different from its source material to become unequivocally “transformative” and whether a court can consider the meaning of the derivative work when it evaluates that transformation. The court’s finding that the Warhol piece is not a fair use could mean trouble for AI-generated works.

所有这些不确定性给使用生成式人工智能的公司带来了一系列挑战。对于供应商和客户对生成式人工智能的使用保持沉默的合同，存在直接或无意侵权的风险。如果企业用户意识到训练数据可能包含未经许可的作品，或者人工智能可以生成不属于合理使用范围的未经授权的衍生作品，那么企业可能会陷入故意侵权的境地，其中每次侵权的损失可能高达 150,000 美元。知用。通过将数据输入生成人工智能工具，还存在意外共享机密商业秘密或商业信息的风险。

All this uncertainty presents a slew of challenges for companies that use generative AI. There are risks regarding infringement—direct or unintentional—in contracts that are silent on generative AI usage by their vendors and customers. If a business user is aware that training data might include unlicensed works or that an AI can generate unauthorized derivative works not covered by fair use, a business could be on the hook for willful infringement, which can include damages up to $150,000 for each instance of knowing use. There’s also the risk of accidentally sharing confidential trade secrets or business information by inputting data into generative AI tools.

降低风险并开辟前进之路

Mitigating Risk and Building a Way Forward

这种新范式意味着公司需要采取新措施来保护自己的短期和长期利益。

This new paradigm means that companies need to take new steps to protect themselves for both the short and long term.

一方面，人工智能开发人员应确保他们在获取用于训练模型的数据时遵守法律。这应该涉及对那些拥有开发人员寻求添加到训练数据中的知识产权的个人进行许可和补偿，无论是通过许可还是分享人工智能工具产生的收入。人工智能工具的客户应询问提供商他们的模型是否接受过任何受保护内容的训练，查看服务条款和隐私政策，并避免使用无法确认其训练数据是否已获得内容创建者的适当许可或受开源约束的生成式人工智能工具人工智能公司遵守的许可证。

AI developers, for one, should ensure that they are in compliance with the law in regard to their acquisition of data being used to train their models. This should involve licensing and compensating those individuals who own the IP that developers seek to add to their training data, whether by licensing it or sharing in revenue generated by the AI tool. Customers of AI tools should ask providers whether their models were trained with any protected content, review the terms of service and privacy policies, and avoid generative AI tools that cannot confirm that their training data is properly licensed from content creators or subject to open-source licenses with which the AI companies comply.

问题诊断

Problem Diagnosis

问题诊断就是确定人工智能要解决的核心问题。换句话说，它涉及识别您希望生成式人工智能实现的主要目标。有些问题相对容易查明，例如当目标是获取特定主题的信息（例如针对员工薪酬的各种人力资源管理策略）时。其他的则更具挑战性，例如探索创新问题的解决方案。

Problem diagnosis is about identifying the core problem for AI to solve. In other words, it concerns identifying the main objective you want generative AI to accomplish. Some problems are relatively simple to pinpoint, such as when the objective is gaining information on a specific topic like various human resources management strategies for employee compensation. Others are more challenging, such as then exploring solutions to an innovation problem.

InnoCentive（现在的 Wazoku Crowd）就是一个典型的例子。该公司已帮助客户解决了 2,500 多个问题，成功率超过 80%，令人印象深刻。我对 InnoCentive 员工的采访表明，这一成功背后的关键因素是他们辨别问题背后根本问题的能力。事实上，他们经常通过使用“五个为什么”技术来区分根本原因和单纯的症状来开始制定问题。

A case in point is InnoCentive (now Wazoku Crowd). The company has helped its clients formulate more than 2,500 problems, with an impressive success rate over 80%. My interviews with InnoCentive employees revealed that a key factor behind this success was their ability to discern the fundamental issue underlying a problem. In fact, they often start their problem formulation process by using the “Five Whys” technique to distinguish the root causes from mere symptoms.

一个特殊的例子是埃克森·瓦尔迪兹灾难性漏油事件后清理亚北极水域的问题。 InnoCentive 与溢油回收研究所合作，查明了石油清理问题的根本原因是原油的粘度：冷冻油变得太稠而无法从驳船上泵出。这一诊断对于最终解决这个长达二十年之久的问题至关重要，该解决方案涉及使用改良版的建筑设备，该设备旨在振动石油，使其保持液态。

A particular instance is the problem of cleaning up subarctic waters after the catastrophic Exxon Valdez oil spill. Collaborating with the Oil Spill Recovery Institute, InnoCentive pinpointed the root cause of the oil cleanup issue as the viscosity of the crude oil: The frozen oil became too thick to pump from barges. This diagnosis was key to finally cracking the two-decade-old problem with a solution that involved using a modified version of construction equipment designed to vibrate the oil, keeping it in a liquid state.

问题分解

Problem Decomposition

问题分解需要将复杂的问题分解为更小的、可管理的子问题。当您处理多方面的问题时，这一点尤其重要，这些问题通常过于复杂，无法生成有用的解决方案。

Problem decomposition entails breaking down complex problems into smaller, manageable subproblems. This is particularly important when you are tackling multifaceted problems, which are often too convoluted to generate useful solutions.

以 InnoCentive 肌萎缩侧索硬化症 (ALS) 挑战为例。挑战不是寻找 ALS 治疗方法这一广泛问题的解决方案，而是集中在其中的一个子组成部分：检测和监测疾病的进展。因此，首次开发了 ALS 生物标记物，提供了一种基于测量流过肌肉组织的电流的无创且经济高效的解决方案。

Take the InnoCentive amyotrophic lateral sclerosis (ALS) challenge, for example. Rather than seeking solutions for the broad problem of discovering a treatment for ALS, the challenge concentrated on a subcomponent of it: detecting and monitoring the progress of the disease. Consequently, an ALS biomarker was developed for the first time, providing a noninvasive and cost-efficient solution based on measuring electrical current flow through muscle tissue.

我使用及时且常见的组织挑战来测试人工智能如何通过问题分解来改进：实施强大的网络安全框架。 Bing 的人工智能解决方案过于广泛和通用，无法立即发挥作用。但在将其分解为子问题（例如安全策略、漏洞评估、身份验证协议和员工培训）后，解决方案得到了显着改进。下面讨论的案例说明了这种差异。方法如因为功能分解或工作分解结构可以帮助您直观地描述复杂问题并简化对与您的组织最相关的各个组件及其互连的识别。

I tested how AI improves with problem decomposition using a timely and common organizational challenge: implementing a robust cybersecurity framework. Bing’s AI-powered solutions were too broad and generic to be immediately useful. But after breaking it down into subproblems—e.g., security policies, vulnerability assessments, authentication protocols, and employee training—the solutions improved considerably. The cases discussed below illustrate the difference. Methods such as functional decomposition or work breakdown structure can help you visually depict complex problems and simplify the identification of individual components and their interconnections that are most relevant for your organization.

问题重构

Problem Reframing

问题重构涉及改变看待问题的视角，从而实现不同的解释。通过以各种方式重构问题，您可以引导人工智能扩大潜在解决方案的范围，从而帮助您找到最佳解决方案并克服创造性障碍。

Problem reframing involves changing the perspective from which a problem is viewed, enabling alternative interpretations. By reframing a problem in various ways, you can guide AI to broaden the scope of potential solutions, which can, in turn, help you find optimal solutions and overcome creative roadblocks.

以 Doug Dietz 为例，他是 GE HealthCare 的创新架构师，他的主要职责是设计最先进的 MRI 扫描仪。在一次去医院时，他看到一名惊恐的孩子正在等待 MRI 扫描，并发现 80% 的儿童需要镇静剂来应对这种令人恐惧的经历。这一发现促使他重新思考了这个问题：“我们如何才能将令人畏惧的 MRI 体验变成孩子们激动人心的冒险？”这一新鲜角度催生了 GE Adventure 系列的开发，该系列将儿科镇静率大幅降低至仅 15%，患者满意度得分提高了 90%，并提高了机器效率。

Consider Doug Dietz, an innovation architect at GE HealthCare, whose main responsibility was designing state-of-the-art MRI scanners. During a hospital visit, he saw a terrified child awaiting an MRI scan and discovered that a staggering 80% of children needed sedation to cope with the intimidating experience. This revelation prompted him to reframe the problem: “How can we turn the daunting MRI experience into an exciting adventure for kids?” This fresh angle led to the development of the GE Adventure Series, which dramatically lowered pediatric sedation rates to a mere 15%, increased patient satisfaction scores by 90%, and improved machine efficiency.

现在想象一下：员工抱怨办公楼缺乏可用停车位。最初的框架可能侧重于增加停车位，但通过从员工的角度重新审视问题（发现停车压力或通勤选择有限），您可以探索不同的解决方案。事实上，当我要求 ChatGPT 使用初始框架和替代框架生成停车位问题的解决方案时，前者产生的解决方案集中于优化停车布局或分配以及寻找新空间。后者提出了多样化的解决方案，例如促进替代交通、可持续通勤和远程工作。

Now imagine this: Employees are complaining about the lack of available parking spaces at the office building. The initial framing may focus on increasing parking space, but by reframing the problem from the employees’ perspective—finding parking stressful or having limited commuting options—you can explore different solutions. Indeed, when I asked ChatGPT to generate solutions for the parking space problem using initial and alternative frames, the former yielded solutions centered on optimizing parking layouts or allocation and finding new spaces. The latter produced a diverse solution set such as promoting alternative transportation, sustainable commuting, and remote work.

为了有效地重构问题，请考虑站在用户的角度，探索类比来表示问题，使用抽象，并主动质疑问题目标或识别问题定义中缺失的组件。

To effectively reframe problems, consider taking the perspective of users, exploring analogies to represent the problem, using abstraction, and proactively questioning problem objectives or identifying missing components in the problem definition.

问题约束设计

Problem Constraint Design

问题约束设计侧重于通过定义输入、过程和过程来描绘问题的边界。解搜索的输出限制。您可以使用约束来指导 AI 生成对当前任务有价值的解决方案。当任务主要以生产力为导向时，采用具体且严格的约束来概述背景、边界和结果标准通常更合适。相比之下，对于以创造力为导向的任务，尝试施加、修改和消除约束可以探索更广阔的解决方案空间并发现新的视角。

Problem constraint design focuses on delineating the boundaries of a problem by defining input, process, and output restrictions of the solution search. You can use constraints to direct AI in generating solutions valuable for the task at hand. When the task is primarily productivity-oriented, employing specific and strict constraints to outline the context, boundaries, and outcome criteria is often more appropriate. In contrast, for creativity-oriented tasks, experimenting with imposing, modifying, and removing constraints allows exploring a wider solution space and discovering novel perspectives.

例如，品牌经理已经在使用 Lately 或 Jasper 等多种人工智能工具来大规模制作有用的社交媒体内容。为了确保这些内容与不同的媒体和品牌形象保持一致，他们通常会对长度、格式、语气或目标受众设置精确的限制。

For example, brand managers are already using several AI tools, such as Lately or Jasper, to produce useful social media content at scale. To ensure this content is aligned with different media and brand image, they are often setting precise constraints on the length, format, tone, or target audience.

然而，在寻求真正的原创性时，品牌经理可以消除格式限制或将输出限制为非常规格式。 GoFundMe 的“帮助改变一切”活动就是一个很好的例子。该公司的目标是制作年度回顾创意内容，不仅能表达对捐助者的感谢并唤起情感，而且能从典型的年终内容中脱颖而出。为了实现这一目标，它设定了非正统的限制：视觉效果将完全依赖人工智能生成的街头壁画风格的艺术，并以所有筹款活动和捐助者为特色。 DALL-E 和稳定版扩散生成单独的图像，然后将其转化为充满情感的视频。结果是：视觉上具有凝聚力和引人注​​目的美感，赢得了广泛的赞誉。²

When seeking true originality, however, brand managers can eliminate formatting constraints or restraining the output to an unconventional format. A great example is GoFundMe’s Help Changes Everything campaign. The company aimed to generate year-in-review creative content that would not only express gratitude to its donors and evoke emotions but also stand out from the typical year-end content. To accomplish this, it set unorthodox constraints: The visuals would rely exclusively on AI-generated street mural–style art and feature all fundraising campaigns and donors. DALL-E and Stable Diffusion generated individual images that were then transformed into an emotionally charged video. The result: a visually cohesive and striking aesthetic that garnered widespread acclaim.²

•••

•  •  •

总体而言，磨练问题诊断、分解、重构和约束设计方面的技能对于使人工智能结果与任务目标保持一致以及促进与人工智能系统的有效协作至关重要。

Overall, honing skills in problem diagnosis, decomposition, reframing, and constraint design is essential for aligning AI outcomes with task objectives and fostering effective collaboration with AI systems.

尽管提示工程可能在短期内成为人们关注的焦点，但其缺乏可持续性、多功能性和可转移性限制了其长期相关性。过分强调完美的文字组合甚至可能会适得其反，因为它可能会分散对问题本身的探索，并削弱用户对创作过程的控制感。相反，掌握问题的表述可能是与复杂的人工智能系统一起驾驭不确定的未来的关键。它可能被证明与计算早期学习编程语言一样重要。

Although prompt engineering may hold the spotlight in the short term, its lack of sustainability, versatility, and transferability limits its long-term relevance. Overemphasizing the crafting of the perfect combination of words can even be counterproductive, as it may detract from the exploration of the problem itself and diminish the user’s sense of control over the creative process. Instead, mastering problem formulation could be the key to navigating the uncertain future alongside sophisticated AI systems. It might prove to be as pivotal as learning programming languages was during the early days of computing.

要点

TAKEAWAYS

尽管围绕提示工程的讨论很热，但它的突出地位可能转瞬即逝。问题表述——识别、分析和描述问题的能力——将是一种更持久、适应性更强的技能，它将继续使我们能够利用生成式人工智能的潜力：

Despite the buzz surrounding prompt engineering, its prominence may be fleeting. Problem formulation—the ability to identify, analyze, and delineate problems—will be a more enduring and adaptable skill that will continue to enable us to harness the potential of generative AI:

✓ 问题表述涉及四个部分：问题诊断、分解、重构和约束设计，并且需要对问题领域有透彻的理解。

✓  Problem formulation involves four components: problem diagnosis, decomposition, reframing, and constraint design—and it necessitates a thorough understanding of the problem domain.

✓ 由于人工智能日益复杂，掌握问题表述可能变得与计算早期学习编程语言一样重要。

✓  Due to the increasing sophistication of AI, mastering problem formulation may become as important as learning programming languages was in the early days of computing.

笔记

NOTES

1. 1 . Thomas Wedell-Wedellsborg，“您解决了正确的问题吗？”，《哈佛商业评论》，2017 年 1 月至 2 月， https:// hbr .org /2017 /01 /are -you -solving -the -right -problems。

2. 1. Thomas Wedell-Wedellsborg, “Are You Solving the Right Problems?,” Harvard Business Review, January–February 2017, https://hbr.org/2017/01/are-you-solving-the-right-problems.

3. 2 . Audrey Kemp，“美国每日广告：GoFundMe 在‘大局观’中描绘捐赠的力量”， Drum，2022 年 12 月 21 日， https:// www .thedrum .com /news /2022 /12 /21 /us -ad -the -day -gofundme -paints -the -power -donating -the -bigger -picture。

4. 2. Audrey Kemp, “US Ad of the Day: GoFundMe Paints the Power of Donating in ‘The Bigger Picture,’ ” Drum, December 21, 2022, https://www.thedrum.com/news/2022/12/21/us-ad-the-day-gofundme-paints-the-power-donating-the-bigger-picture.

改编自 hbr.org 上发布的内容，2023 年 6 月 6 日（产品#H07NQK）。

Adapted from content posted on hbr.org, June 6, 2023 (product #H07NQK).

1. 我应该如何准备在我的组织中引入人工智能？

1. How should I prepare to introduce AI at my organization?

首先，重要的是要认识到使用人工智能的最佳方式不同于我们使用其他新技术的方式。过去，大多数新工具只是让我们能够更有效地执行任务。人们用笔书写，然后是打字机（速度更快），然后是计算机（速度更快）。每个新工具都可以提高写作效率，但一般流程（起草、修改、编辑）基本保持不变。

To start, it’s important to recognize that the optimal way to work with AI is different from the way we’ve worked with other new technologies. In the past, most new tools simply enabled us to perform tasks more efficiently. People wrote with pens, then typewriters (which were faster), then computers (which were even faster). Each new tool allowed for more efficient writing, but the general processes (drafting, revising, editing) remained largely the same.

人工智能则不同。它对我们的工作和流程有更实质性的影响，因为它能够找到我们看不到的模式，然后使用它们来提供见解和分析、预测、建议，甚至完整的草稿。因此，我们不应将人工智能视为我们使用的工具，而应将其视为一组可以与之协作的系统。

AI is different. It has a more substantial influence on our work and our processes because it’s able to find patterns that we can’t see and then use them to provide insights and analysis, predictions, suggestions, and even full drafts all on its own. So instead of thinking of AI as the tools we use, we should think of it as a set of systems with which we can collaborate.

要在组织中与人工智能进行有效协作，请重点关注三件事：

To effectively collaborate with AI at your organization, focus on three things:

2. 如何确保人工智能决策的透明度？

2. How can we ensure transparency in how AI makes decisions?

领导者需要认识到，并不总是能够了解人工智能系统如何做出决策。人工智能能够快速处理大量数据并比人类更准确或更高效地执行某些任务的一些特征也可能使其成为黑匣子：我们无法看到输出是如何产生的。然而，我们都可以通过两种方式在提高人工智能决策过程的透明度和问责制方面发挥作用：

Leaders need to recognize that it is not always possible to know how AI systems are making decisions. Some of the very characteristics that allow AI to quickly process huge amounts of data and perform certain tasks more accurately or efficiently than humans can also make it a black box: We can’t see how the output was produced. However, we can all play a role in increasing transparency and accountability in AI decision-making processes in two ways:

3. 我们如何在法学硕士周围竖起护栏，使他们的回答真实并符合我们想要塑造的品牌形象？

3. How can we erect guardrails around LLMs so that their responses are true and consistent with the brand image we want to project?

法学硕士存在一些严重的风险。他们能：

LLMs come with several serious risks. They can:

• 通过采用负面刻板印象或最小化少数群体观点来维持有害偏见
• Perpetuate harmful bias by deploying negative stereotypes or minimizing minority viewpoints
• 通过重复谎言或编造事实和引用来传播错误信息
• Spread misinformation by repeating falsehoods or making up facts and citations
• 未经他人同意使用数据侵犯隐私
• Violate privacy by using data without people’s consent
• 如果它们被用来生成网络钓鱼电子邮件或其他网络攻击，则会导致安全漏洞
• Cause security breaches if they are used to generate phishing emails or other cyberattacks
• 由于训练和运行它们需要大量的计算资源，因此会损害环境
• Harm the environment because of the significant computational resources required to train and run them

数据管理和文档记录是减少这些风险并确保法学硕士给出更符合（而不损害）您的品牌形象的回应的两种方法。

Data curation and documentation are two ways to curtail those risks and ensure that LLMs will give responses that are more consistent with—not harmful to—your brand image.

4. 如何确保用于训练人工智能模型的数据集具有代表性且不包含有害偏差？

4. How can we ensure that the dataset we use to train AI models is representative and doesn’t include harmful biases?

清理数据集是一项挑战，您的组织可以通过优先考虑透明度和公平性而不是模型大小以及在数据管理中代表不同的人群来帮助克服。

Sanitizing datasets is a challenge that your organization can help overcome by prioritizing transparency and fairness over model size and by representing diverse populations in data curation.

首先，考虑您所做的权衡。科技公司一直在追求更大的人工智能系统，因为它们在某些任务上往往更有效，比如维持人类的对话。然而，如果模型太大而无法完全理解，就不可能消除潜在的偏差。为了充分消除有害偏见，开发人员必须能够理解并记录数据集固有的风险，这可能意味着使用较小的数据集。

First, consider the trade-offs you make. Tech companies have been pursuing larger AI systems because they tend to be more effective at certain tasks, like sustaining human-seeming conversations. However, if a model is too large to fully understand, it’s impossible to rid it of potential biases. To fully combat harmful bias, developers must be able to understand and document the risks inherent to a dataset, which might mean using a smaller one.

其次，如果多元化的团队（包括代表性不足的群体的成员）收集并生成用于训练模型的数据，那么您将有更好的机会确保具有不同观点和认同的人它们代表了实体。这种做法还有助于识别数据中未被识别的偏差或盲目性。

Second, if diverse teams, including members of underrepresented populations, collect and produce the data used to train models, you’ll have a better chance of ensuring that people with a variety of perspectives and identities are represented in them. This practice also helps identify unrecognized biases or blinders in the data.

人工智能只有在公平运行时才会值得信赖，而只有当我们优先考虑数据和开发团队的多样化，并清楚地记录人工智能是如何设计以实现公平时，这种情况才会发生。

AI will only be trustworthy once it works equitably, and that will happen only if we prioritize diversifying data and development teams and clearly document how AI has been designed for fairness.

5.人工智能侵犯数据隐私的潜在风险有哪些？

5. What are the potential risks of data privacy violations with AI?

使用敏感员工和客户数据的人工智能很容易受到不良行为者的攻击。为了应对这些风险，组织应该尽可能多地了解人工智能的开发方式，然后决定是否适合使用安全数据。他们还应该保持技术系统更新并指定预算资源以确保软件安全。这需要持续采取行动，因为一个小漏洞可能会让整个组织面临漏洞。

AI that uses sensitive employee and customer data is vulnerable to bad actors. To combat these risks, organizations should learn as much as they can about how their AI has been developed and then decide whether it’s appropriate to use secure data with it. They should also keep tech systems updated and earmark budget resources to keep the software secure. This requires continuous action, as a small vulnerability can leave an entire organization open to breaches.

区块链创新可以在这方面提供帮助。区块链是一种记录数据交易的安全的分布式账本，目前用于创建支付系统（更不用说加密货币）等应用程序。

Blockchain innovations can help on this front. A blockchain is a secure, distributed ledger that records data transactions, and it’s currently being used for applications like creating payment systems (not to mention cryptocurrencies).

当涉及更广泛的运营时，请考虑安大略省前信息和隐私专员 Ann Cavoukian 提出的隐私设计 (PbD) 框架，该框架建议组织遵循七项基本原则：

When it comes to your operations more broadly, consider this privacy by design (PbD) framework from former information and privacy commissioner of Ontario Ann Cavoukian, which recommends that organizations embrace seven foundational principles:

• 积极主动，而非被动反应——预防性而非补救性。
• Be proactive, not reactive—preventive, not remedial.
• 以隐私作为默认设置。
• Lead with privacy as the default setting.
• 将隐私融入设计中。
• Embed privacy into design.
• 保留完整功能，包括隐私和安全。
• Retain full functionality, including privacy and security.
• 确保端到端安全。
• Ensure end-to-end security.
• 保持可见性和透明度。
• Maintain visibility and transparency.
• 尊重用户隐私——保持系统以用户为中心。⁸
• Respect user privacy—keep systems user-centric.⁸

将 PbD 原则纳入您的运营需要的不仅仅是雇用隐私人员或创建隐私部门。您组织中的所有人员都需要了解客户和员工对这些问题的担忧。隐私不是事后才想到的；而是事后才想到的。它需要成为数字化运营的核心，每个人都需要努力保护它。

Incorporating PbD principles into your operation requires more than hiring privacy personnel or creating a privacy division. All the people in your organization need to be attuned to customer and employee concerns about these issues. Privacy isn’t an afterthought; it needs to be at the core of digital operations, and everyone needs to work to protect it.

6. 我们如何鼓励员工使用人工智能来提高生产力，而不是简单地走捷径？

6. How can we encourage employees to use AI for productivity purposes and not simply to take shortcuts?

即使法学硕士出现了，人工智能技术还无法执行人类可以执行的一系列令人眼花缭乱的任务，而且有很多事情它比普通人做得更糟糕。有效地使用每个新工具需要了解其目的。

Even with the advent of LLMs, AI technology is not yet capable of performing the dizzying range of tasks that humans can, and there are many things that it does worse than the average person. Using each new tool effectively requires understanding its purpose.

例如，考虑一下 ChatGPT。通过学习语言模式，它已经变得非常擅长预测哪些单词应该跟随其他单词，以至于它可以对复杂的问题产生看似复杂的文本响应。然而，这些输出的质量是有限的，因为善于猜测单词和短语的合理组合与理解材料不同。因此，ChatGPT 可以创作一首莎士比亚风格的诗歌，因为它已经了解了莎士比亚的戏剧和诗歌的特定模式，但它无法产生对他的作品中的人类状况的原始洞察。

For example, think about ChatGPT. By learning about language patterns, it has become so good at predicting which words are supposed to follow others that it can produce seemingly sophisticated text responses to complicated questions. However, there’s a limit to the quality of these outputs because being good at guessing plausible combinations of words and phrases is different from understanding the material. So ChatGPT can produce a poem in the style of Shakespeare because it has learned the particular patterns of his plays and poems, but it cannot produce the original insight into the human condition that informs his work.

相比之下，人工智能在做出预测方面可以比人类更好、更高效，因为它可以更快地处理大量数据。例子包括根据言语模式预测早期痴呆症，检测人眼无法区分的癌性肿瘤，并规划战场上更安全的路线。

By contrast, AI can be better and more efficient than humans at making predictions because it can process much larger amounts of data much more quickly. Examples include predicting early dementia from speech patterns, detecting cancerous tumors indistinguishable to the human eye, and planning safer routes through battlefields.

因此，应该鼓励员工评估人工智能的优势是否与任务相匹配，并采取相应的行动。如果您需要快速处理大量信息，它可以做到。如果你需要一堆新想法，它可以产生它们。即使您需要做出困难的决定，只要经过相关数据的培训，它也可以提供建议。

Employees should therefore be encouraged to evaluate whether AI’s strengths match up to a task and proceed accordingly. If you need to process a lot of information quickly, it can do that. If you need a bunch of new ideas, it can generate them. Even if you need to make a difficult decision, it can offer advice, providing it’s been trained on relevant data.

但你不应该在没有人类监督的情况下使用人工智能来创建有意义的工作产品。如果您需要编写大量内容非常相似的文档，人工智能可能是长期以来被称为“样板”材料的有用生成器。但请注意，它的输出来自其数据集和算法，并且它们不一定是好的或准确的。

But you shouldn’t use AI to create meaningful work products without human oversight. If you need to write a quantity of documents with very similar content, AI may be a useful generator of what has long been referred to as “boilerplate” material. But be aware that its outputs are derived from its datasets and algorithms, and they aren’t necessarily good or accurate.

7. 我们应该有多担心人工智能会取代工作？

7. How worried should we be that AI will replace jobs?

每一次技术革命创造的就业机会都多于它摧毁的就业机会。汽车让马车司机失业，但却带来了建造和修理汽车、经营加油站等新的工作岗位。人工智能技术的新颖性让人很容易担心它们会取代劳动力中的人类。但我们应该将它们视为增强人类表现的方法。例如，像 Collective[i] 这样的公司开发了人工智能系统，可以分析数据以快速生成高度准确的销售预测；传统上，这项工作需要人们几天或几周的时间才能完成。但没有销售人员失业。相反，他们有更多的时间专注于工作中更重要的部分：建立关系、管理和实际销售。

Every technological revolution has created more jobs than it has destroyed. Automobiles put horse-and-buggy drivers out of business but led to new jobs building and fixing cars, running gas stations, and more. The novelty of AI technologies makes it easy to fear they will replace humans in the workforce. But we should instead view them as ways to augment human performance. For example, companies like Collective[i] have developed AI systems that analyze data to produce highly accurate sales forecasts quickly; traditionally, this work took people days or weeks to pull together. But no salespeople are losing their jobs. Rather, they’ve got more time to focus on more important parts of their work: building relationships, managing, and actually selling.

同样，OpenAI 的 Codex 等服务可以自动生成用于基本目的的编程代码。这并没有取代程序员；而是取代了程序员。它使他们能够更有效地编写代码并自动执行测试等重复任务，以便他们可以处理更高级别的问题，例如系统架构、领域建模和用户体验。

Similarly, services like OpenAI’s Codex can autogenerate programming code for basic purposes. This doesn’t replace programmers; it allows them to write code more efficiently and automate repetitive tasks like testing so that they can work on higher-level issues such as systems architecture, domain modeling, and user experience.

对就业的长期影响是复杂且不平衡的，某些行业或地区可能会出现就业被破坏和取代的时期。为了确保技术进步的好处被广泛分享，投资于教育和劳动力发展以帮助人们适应新的就业市场至关重要。

The long-term effects on jobs are complex and uneven, and there can be periods of job destruction and displacement in certain industries or regions. To ensure that the benefits of technological progress are widely shared, it is crucial to invest in education and workforce development to help people adapt to the new job market.

个人和组织应重点关注技能提升和扩展，为充分利用新技术做好准备。人工智能和机器人不会很快取代人类。更有可能的现实是，具有数字思维的人将取代那些没有数字思维的人。

Individuals and organizations should focus on upskilling and scaling to prepare to make the most of new technologies. AI and robots aren’t replacing humans anytime soon. The more likely reality is that people with digital mindsets will replace those without them.

8. 我的组织如何确保我们开发或使用的人工智能不会伤害个人或团体或侵犯人权？

8. How can my organization ensure that the AI we develop or use won’t harm individuals or groups or violate human rights?

人工智能偏见的危害已被广泛记录。 Joy Buolamwini 和 Timnit Gebru 在 2018 年的开创性论文《性别阴影》中表明，IBM 和微软等公司提供的流行面部识别技术在识别白人男性面孔方面近乎完美，但错误识别黑人女性面孔的几率高达 35%。⁹面部识别可用于解锁手机，但也可用于监控麦迪逊广场花园的顾客、监视抗议者以及警方调查中的嫌疑人，而错误识别会导致错误逮捕，从而扰乱人们的生活。随着人工智能的力量不断增强并越来越融入我们的日常生活，其潜在危害也呈指数级增长。以下是保护人工智能的策略。

The harms of AI bias have been widely documented. In their seminal 2018 paper “Gender Shades,” Joy Buolamwini and Timnit Gebru showed that popular facial recognition technologies offered by companies like IBM and Microsoft were nearly perfect at identifying white male faces but misidentified Black female faces as much as 35% of the time.⁹ Facial recognition can be used to unlock your phone but is also used to monitor patrons at Madison Square Garden, surveil protesters, and tap suspects in police investigations—and misidentification has led to wrongful arrests that can derail people’s lives. As AI grows in power and becomes more integrated into our daily lives, its potential for harm grows exponentially, too. Here are strategies to safeguard AI.

笔记

NOTES

1. 1 . Tsedal Neely，“遵循 30% 规则培养数字化思维”，LinkedIn，2022 年 5 月 12 日， https:// www .linkedin .com /pulse /developing -digital -mindset -following -30 -rule -tsedal -neeley /。

2. 1. Tsedal Neely, “Developing a Digital Mindset by Following the 30% Rule,” LinkedIn, May 12, 2022, https://www.linkedin.com/pulse/developing-digital-mindset-following-30-rule-tsedal-neeley/.

3. 2 . Callen Anthony、Beth A. Bechky 和 ​​Anne-Laure Fayard，“与人工智能‘协作’：以系统视角探索工作的未来”，《组织科学》，2023 年 1 月 9 日， https:// pubsonline .informs .org /doi /abs /10 .1287 /orsc .2022 .1651 ?journalCode =orsc。

4. 2. Callen Anthony, Beth A. Bechky, and Anne-Laure Fayard, “ ‘Collaborating’ with AI: Taking a System View to Explore the Future of Work,” Organization Science, January 9, 2023, https://pubsonline.informs.org/doi/abs/10.1287/orsc.2022.1651?journalCode=orsc.

5. 3 . Thomas W. Malone、Daniela Rus 和 Robert Laubacher，“人工智能与工作的未来”，麻省理工学院研究简报 17（2020 年 12 月）， https:// workofthefuture .mit .edu /wp -content /uploads /2020 /12 /2020 -研究-简报-Malone -Rus -Laubacher2 .pdf。

6. 3. Thomas W. Malone, Daniela Rus, and Robert Laubacher, “Artificial Intelligence and the Future of Work,” MIT Research Brief 17 (December 2020), https://workofthefuture.mit.edu/wp-content/uploads/2020/12/2020-Research-Brief-Malone-Rus-Laubacher2.pdf.

7. 4 . Emily M. Bender、Timnit Gebru、Angelina McMillan-Major 和 Shmargaret Shmitchell，“随机鹦鹉的危险：语言模型会太大吗？”， 2021 年 ACM 公平、问责和透明度会议记录，FAccT 21 日，纽约州纽约市，2021 年 3 月， https:// dl .acm .org /doi /10 .1145 /3442188 .3445922。

8. 4. Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell, “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?,” Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’21, New York, NY, March 2021, https://dl.acm.org/doi/10.1145/3442188.3445922.

9. 5 . Robert M. Bond 和 R. Kelly Garrett，“参与 Reddit 上经过事实核查的帖子”， PNAS Nexus 2，第 1 期。 3 （2023年3月）， https : //academic.oup.com/pnasnexus/article/2/3/pgad018/7008465 。

10. 5. Robert M. Bond and R. Kelly Garrett, “Engagement with Fact-Checked Posts on Reddit,” PNAS Nexus 2, no. 3 (March 2023), https://academic.oup.com/pnasnexus/article/2/3/pgad018/7008465.

11. 6 . “ 2021年社区洞察报告，蓬勃发展的运动” ，维基媒体Meta - Wiki ，https : //meta.m.wikimedia.org/wiki/Community_Insights/Community_Insights_2021_Report/Thriving_Movement#Community_and_Newcomer_Diversity。

12. 6. “Community Insights 2021 Report, Thriving Movement,” Wikimedia Meta-Wiki, https://meta.m.wikimedia.org/wiki/Community_Insights/Community_Insights_2021_Report/Thriving_Movement#Community_and_Newcomer_Diversity.

13. 7 . Bender 等人，“论随机鹦鹉的危险”。

14. 7. Bender et al., “On the Dangers of Stochastic Parrots.”

15. 8 . Ann Cavoukian，“隐私设计：7 项基本原则”，privacybydesign.ca，2011 年 1 月， https:// www .ipc .on .ca /wp -content /uploads /Resources /7foundationalprinciples .pdf。

16. 8. Ann Cavoukian, “Privacy by Design: the 7 Foundational Principles,” privacybydesign.ca, January 2011, https://www.ipc.on.ca/wp-content/uploads/Resources/7foundationalprinciples.pdf.

17. 9 . Joy Buolamwini 和 Timnit Gebru，“性别阴影：商业性别分类中的交叉准确性差异”，第一届公平、问责和透明度会议记录， PMLR 81 (2018)：77–91， https: //processing.mlr.press /v81 /buolamwini18a .html。

18. 9. Joy Buolamwini and Timnit Gebru, “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification,” Proceedings of the 1st Conference on Fairness, Accountability and Transparency, PMLR 81 (2018): 77–91, https://proceedings.mlr.press/v81/buolamwini18a.html.

19. 10 . Rebecca Bellan，“微软在加倍投入 OpenAI 的同时解雇了道德 AI 团队”，TechCrunch，2023 年 3 月 13 日， https:// techcrunch .com /2023 /03 /13 /microsoft -lays -off -an -ethical - ai -team -as -it -doubles -down -on -openai / .

20. 10. Rebecca Bellan, “Microsoft Lays Off an Ethical AI Team as It Doubles Down on OpenAI,” TechCrunch, March 13, 2023, https://techcrunch.com/2023/03/13/microsoft-lays-off-an-ethical-ai-team-as-it-doubles-down-on-openai/.

改编自 hbr.org 上发布的内容，2023 年 5 月 9 日（产品#H07MEI）。

Adapted from content posted on hbr.org, May 9, 2023 (product #H07MEI).

生成人工智能的道德发展指南

Guidelines for the Ethical Development of Generative AI

随着这些工具获得主流采用，我们的一套新指南可以帮助组织评估生成式人工智能的风险和注意事项。它们涵盖五个重点领域。

Our new set of guidelines can help organizations evaluate generative AI’s risks and considerations as these tools gain mainstream adoption. They cover five focus areas.

集成生成式人工智能

Integrating Generative AI

大多数组织将集成生成式人工智能工具，而不是构建自己的工具。以下是一些将生成式人工智能安全地集成到业务应用程序中以推动业务成果的战术技巧：

Most organizations will integrate generative AI tools rather than build their own. Here are some tactical tips for safely integrating generative AI in business applications to drive business results:

笔记

NOTES

1. 1 .生成式 AI 为‘游戏规则改变者’ ，但寻求道德和信任方面的进步”，salesforce.com，2023 年 3 月 6 日， https : //www.salesforce.com/news/stories/generative-ai-research/ 。

2. 1. “IT Leaders Call Generative AI a ‘Game Changer’ but Seek Progress on Ethics and Trust,” salesforce.com, March 6, 2023, https://www.salesforce.com/news/stories/generative-ai-research/.

3. 2 . “Salesforce 首次推出 AI 道德模型：如何道德实践进一步负责任的人工智能”， salesforce.com，2021 年 9 月 2 日， https:// www .salesforce .com /news /stories /salesforce -debuts -ai -ethics -model -how -道德-实践-进一步-负责任-人工智能/。

4. 2. “Salesforce Debuts AI Ethics Model: How Ethical Practices Further Responsible Artificial Intelligence,” salesforce.com, September 2, 2021, https://www.salesforce.com/news/stories/salesforce-debuts-ai-ethics-model-how-ethical-practices-further-responsible-artificial-intelligence/.

5. 3 . Saul 和 Dina Bass，“人工智能正在蓬勃发展——它的碳足迹也是如此”，彭博社，2023年3月9日， https : //www.bloomberg.com/news/articles/2023-03-09/how-much-energy -do -ai -和-chatgpt -use -no -one -knows -for -sure； Nabiha Syed，“人工智能技术的秘密水足迹：与 Shaolei Ren的对话” ， Markup，2023 年 4 月 15日，https : //themarkup.org/hello-world/2023/04/15/the-secret-water人工智能技术的足迹。

6. 3. Saul and Dina Bass, “Artificial Intelligence Is Booming—So Is Its Carbon Footprint,” Bloomberg, March 9, 2023, https://www.bloomberg.com/news/articles/2023-03-09/how-much-energy-do-ai-and-chatgpt-use-no-one-knows-for-sure; Nabiha Syed, “The Secret Water Footprint of AI Technology: A Conversation with Shaolei Ren,” The Markup, April 15, 2023, https://themarkup.org/hello-world/2023/04/15/the-secret-water-footprint-of-ai-technology.

改编自 hbr.org 上发布的内容，2023 年 6 月 5 日（产品#H07OC4）。

Adapted from content posted on hbr.org, June 5, 2023 (product #H07OC4).

“人工智能”到底意味着什么？

What Does “AI” Actually Mean?

“‘人工智能驱动’是科技领域毫无意义的‘纯天然’。 ”

“ ‘AI-powered’ is tech’s meaningless equivalent of ‘all natural.’ ”

—Devin Coldewey， TechCrunch，2022 年

—Devin Coldewey, TechCrunch, 2022

人工智能无法摆脱通用人工智能有两个原因。首先，这个术语通常被随意使用，而没有明确我们是在谈论 AGI 还是狭义 AI（一个术语）这本质上意味着实用、专注的机器学习部署。尽管存在巨大差异，但它们之间的界限在常见的言论和软件销售材料中却很模糊。

AI cannot get away from AGI for two reasons. First, the term is generally thrown around without clarifying whether we’re talking about AGI or narrow AI, a term that essentially means practical, focused ML deployments. Despite the tremendous differences, the boundary between them blurs in common rhetoric and software sales materials.

其次，除了AGI之外，还没有令人满意的方式来定义AI。将人工智能定义为 AGI 之外的事物本身已成为一项研究挑战，尽管这是一个堂吉诃德式的挑战。如果它不意味着 AGI，那就没有任何意义——其他建议的定义要么无法符合“人工智能”所隐含的雄心勃勃的精神中的“智能”，要么无法建立客观的目标。我们面临着这个难题，无论是试图确定（1）人工智能的定义，（2）计算机“智能”的标准，还是（3）证明真正人工智能的性能基准。这三者是一回事。

Second, there’s no satisfactory way to define AI besides AGI. Defining AI as something other than AGI has become a research challenge unto itself, albeit a quixotic one. If it doesn’t mean AGI, it doesn’t mean anything—other suggested definitions either fail to qualify as “intelligent” in the ambitious spirit implied by “AI” or fail to establish an objective goal. We face this conundrum whether trying to pinpoint (1) a definition for AI, (2) the criteria by which a computer would qualify as “intelligent,” or (3) a performance benchmark that would certify true AI. These three are one and the same.

问题出在“智力”这个词本身。当用来描述一台机器时，它是极其模糊的。如果人工智能本来就是一个合法的领域，那么这就是个坏消息。工程不能追求不精确的目标。如果无法定义它，就无法构建它。为了开发一个装置，你必须能够衡量它有多好——它的表现如何以及你离目标有多近——这样你就知道你正在取得进展，并且最终知道你何时取得了成功。开发它。

The problem is with the word intelligence itself. When used to describe a machine, it’s relentlessly nebulous. That’s bad news if AI is meant to be a legitimate field. Engineering can’t pursue an imprecise goal. If you can’t define it, you can’t build it. To develop an apparatus, you must be able to measure how good it is—how well it performs and how close you are to the goal—so that you know you’re making progress and so that you ultimately know when you’ve succeeded in developing it.

为了摆脱这种困境，该行业不断地表演着一场尴尬的人工智能定义之舞，但徒劳无功。我称之为AI 洗牌。人工智能意味着计算机可以智能地完成某些事情（循环定义）。不，它是机器展示的智能（如果可能的话，甚至更循环）。相反，它是一个采用某些先进方法的系统，例如机器学习、自然语言处理、基于规则的系统、语音识别、计算机视觉或其他以概率方式运行的技术（显然，采用这些方法中的一种或多种并不自动使系统具有智能性）。

In a vain attempt to fend off this dilemma, the industry continually performs an awkward dance of AI definitions that I call the AI shuffle. AI means computers that do something smart (a circular definition). No, it’s intelligence demonstrated by machines (even more circular, if that’s possible). Rather, it’s a system that employs certain advanced methodologies, such as ML, natural language processing, rule-based systems, speech recognition, computer vision, or other techniques that operate probabilistically (clearly, employing one or more of these methods doesn’t automatically qualify a system as intelligent).

但是，如果一台机器看起来足够像人类，如果你无法通过在聊天室中询问它（著名的图灵测试）来区分它和人类，那么它肯定有资格成为智能机器。但愚弄人们的能力是一个任意的、不断变化的目标，因为随着时间的推移，人类受试者会变得更加聪明。任何给定的系统最多只能通过一次测试——愚弄我们两次，这是人类的耻辱。通过图灵测试未能达到目标的另一个原因是愚弄人们的价值或效用有限。如果人工智能能够存在，它肯定会有用。

But surely a machine would qualify as intelligent if it seemed sufficiently humanlike, if you couldn’t distinguish it from a human, say, by interrogating it in a chatroom—the famous Turing test. But the ability to fool people is an arbitrary, moving target, since human subjects become wiser to the trickery over time. Any given system will only pass the test at most once—fool us twice, shame on humanity. Another reason that passing the Turing test misses the mark is because there’s limited value or utility in fooling people. If AI could exist, certainly it’s supposed to be useful.

如果我们根据人工智能的能力来定义它呢？例如，如果我们将人工智能定义为可以执行传统上需要人类完成的困难任务的软件，例如驾驶汽车、掌握国际象棋或识别人脸。事实证明这个定义也不起作用，因为一旦计算机可以做某事，我们倾向于轻视它。毕竟，计算机只能管理易于理解和明确指定的机械任务。一旦克服了挑战，成就就会突然失去魅力，而能够做到这一点的计算机似乎并不“智能”——至少没有达到人工智能一词所期望的全心全意的程度。一旦计算机掌握了国际象棋，我们就几乎没有感觉我们已经“解决”了人工智能。

What if we define AI by what it’s capable of? For example, if we define AI as software that can perform a task so difficult that it traditionally requires a human, such as driving a car, mastering chess, or recognizing human faces. It turns out that this definition doesn’t work either because, once a computer can do something, we tend to trivialize it. After all, computers can manage only mechanical tasks that are well understood and well specified. Once the challenge is surmounted, the accomplishment suddenly loses its charm and the computer that can do it doesn’t seem “intelligent” after all—at least not to the wholehearted extent intended by the term AI. Once computers mastered chess, there was little feeling that we’d “solved” AI.

这个悖论被称为人工智能效应，它告诉我们，如果可能的话，它就不是智能的。由于目标始终难以捉摸，人工智能无意中等同于“让计算机做计算机难以做到的事情”——人工不可能。一旦到达，任何目的地都不会令你满意；人工智能完全违背了定义。具有讽刺意味的是，计算机科学先驱拉里·特斯勒（Larry Tesler）提出了一句著名的建议，我们不妨将人工智能定义为“机器尚未完成的事情”。

This paradox, known as the AI effect, tells us that if it’s possible, it’s not intelligent. Suffering from an ever-elusive objective, AI inadvertently equates to “getting computers to do things too difficult for computers to do”—artificial impossibility. No destination will satisfy once you arrive; AI categorically defies definition. With due irony, the computer science pioneer Larry Tesler famously suggested that we might as well define AI as “whatever machines haven’t done yet.”

讽刺的是，正是机器学习的巨大成功首先推动了人工智能的发展。毕竟，简而言之，提高可衡量的性能就是监督机器学习。根据基准（例如标记数据样本）评估系统的反馈将指导其下一步改进。通过这个过程，机器学习以无数的方式提供了前所未有的价值。它被誉为“我们最重要的通用技术”正如安德鲁·麦卡菲和埃里克·布林约尔松所说的那样。²最重要的是，机器学习的飞跃性发展推动了人工智能的炒作。

Ironically, it was ML’s measurable success that hyped up AI in the first place. After all, improving measurable performance is supervised machine learning in a nutshell. The feedback from evaluating the system against a benchmark—such as a sample of labeled data—guides its next improvement. By this process, ML delivers unprecedented value in countless ways. It has earned its title as “the most important general-purpose technology of our era,” as Andrew McAfee and Erik Brynjolfsson put it.² More than anything else, ML’s proven leaps and bounds have fueled AI hype.

全面采用通用人工智能

All In with Artificial General Intelligence

“我预测未来五年内我们将迎来第三个人工智能冬天。……当我在 91 年获得人工智能和机器学习博士学位时，人工智能实际上是一个坏词。没有公司会考虑雇用人工智能领域的人。”

“I predict we will see the third AI winter within the next five years.… When I graduated with my PhD in AI and ML in ’91, AI was literally a bad word. No company would consider hiring somebody who was in AI.”

—Usama Fayyad，2022 年 6 月 23 日，在机器学习周上发言

—Usama Fayyad, June 23, 2022, speaking at Machine Learning Week

有一种方法可以克服这种定义困境：全力以赴，将“AI”定义为AGI，即能够完成人类可以完成的任何智力任务的软件。如果这个听起来像科幻小说的目标得以实现，我认为将会有一个强有力的论据证明它具有“智能”。这是一个可衡量的目标——至少在原则上是这样，如果不是在实践中的话。例如，它的开发人员可以针对一组 100 万个任务对系统进行基准测试，其中包括您可能发送给虚拟助理的数万个复杂的电子邮件请求、您可能向机器人发出的针对仓库员工的各种指令，甚至用一段简短的概述来说明机器作为首席执行官应该如何运营一家财富500 强公司以实现盈利。

There is one way to overcome this definition dilemma: Go all in and define “AI” as AGI, software capable of any intellectual task humans can do. If this science fiction–sounding goal were achieved, I submit that there would be a strong argument that it qualified as “intelligent.” And it’s a measurable goal—at least in principle, if not in practicality. For example, its developers could benchmark the system against a set of 1 million tasks, including tens of thousands of complicated email requests you might send to a virtual assistant, various instructions for a warehouse employee you’d just as well issue to a robot, and even brief one-paragraph overviews for how the machine should, in the role of CEO, run a Fortune 500 company to profitability.

AGI 可能设定了一个明确的目标，但它是超凡脱俗的——一个难以实现的野心。没有人知道它是否以及何时能够实现。

AGI may set a clear-cut objective, but it’s out of this world—as unwieldy an ambition as there can be. Nobody knows if and when it could be achieved.

这就是典型机器学习项目的问题。通过将它们称为“人工智能”，我们​​表达了它们与 AGI 处​​于同一范围，它们是建立在积极朝这个方向发展的技术之上的。 “AI”困扰着机器学习。它引发了宏大的叙事并激发了人们的期望，以不切实际的方式出售真正的技术。这让决策者和陷入困境的项目感到困惑。

Therein lies the problem for typical ML projects. By calling them “AI,” we convey that they sit on the same spectrum as AGI, that they’re built on technology that is actively inching along in that direction. “AI” haunts ML. It invokes a grandiose narrative and pumps up expectations, selling real technology in unrealistic terms. This confuses decision-makers and dead-end projects left and right.

如果 AI 的成分与 AGI 相同，那么很多人都想分一杯羹，这是可以理解的。 AGI 所承诺的愿望实现——一种终极力量——如此诱人，几乎无法抗拒。

It’s understandable that so many would want to claim a piece of the AI pie, if it’s made of the same ingredients as AGI. The wish fulfillment AGI promises—a kind of ultimate power—is so seductive that it’s nearly irresistible.

但有一种更好的前进方式，一种现实的方式，我认为它已经足够令人兴奋了：更有效地运行主要业务——我们作为组织所做的主要事情！大多数商业机器学习项目的目标就是做到这一点。为了让他们以更高的速度取得成功，我们必须脚踏实地。如果您的目标是提供运营价值，请不要购买“人工智能”，也不要出售“人工智能”。说出您的意思并表达您的意思。如果一项技术由机器学习组成，我们就这样称呼它。

But there’s a better way forward, one that’s realistic and that I would argue is already exciting enough: running major operations—the main things we do as organizations—more effectively! Most commercial ML projects aim to do just that. For them to succeed at a higher rate, we’ve got to come down to earth. If your aim is to deliver operational value, don’t buy “AI” and don’t sell “AI.” Say what you mean and mean what you say. If a technology consists of ML, let’s call it that.

关于人类思维即将过时的报道被严重夸大了，这意味着人工智能幻灭的另一个时代即将到来。而且，从长远来看，只要我们继续夸张地应用人工智能这个术语，我们就将继续经历人工智能的冬天。但如果我们淡化言辞——或者以其他方式将机器学习与人工智能区分开来——我们就能正确地将机器学习作为一个行业与下一个人工智能冬天隔离开来。这包括抵制诱惑，不要随波逐流，不要被动地肯定那些似乎在全能人工智能的祭坛上低头的不切实际的决策者。否则，危险就显而易见：当炒作消退、过度销售被揭穿、冬天到来时，机器学习的大部分真正价值主张将与神话一起被不必要地抛弃，就像婴儿被洗澡水一样。

Reports of the human mind’s looming obsolescence have been greatly exaggerated, which means another era of AI disillusionment is nigh. And, in the long run, we will continue to experience AI winters as long as we continue to hyperbolically apply the term AI. But if we tone down the rhetoric—or otherwise differentiate ML from AI—we will properly insulate ML as an industry from the next AI winter. This includes resisting the temptation to ride hype waves and refrain from passively affirming starry-eyed decision-makers who appear to be bowing at the altar of an all-capable AI. Otherwise, the danger is clear and present: When the hype fades, the overselling is debunked, and winter arrives, much of ML’s true value proposition will be unnecessarily disposed of along with the myths, like the baby with the bathwater.

要点

TAKEAWAYS

随着生成式人工智能每隔几个月就会发布令人惊叹的新功能，并且人工智能炒作以更高的速度升级，现在是我们将当今大多数实用机器学习 (ML) 项目与生成式人工智能的进步区分开来的时候了。

With breathtaking new capabilities from generative AI released every several months—and AI hype escalating at an even higher rate—it’s high time we differentiate most of today’s practical machine learning (ML) projects from generative AI’s advances.

✓ 对于大多数机器学习项目来说， “人工智能”这个词太过分了。它指的是人类水平的能力，更好地描述为 AGI（通用人工智能）——能够完成人类可以完成的任何智力任务的软件——而且没有人知道 AGI 是否以及何时能够实现。

✓  For most ML projects, the term AI goes entirely too far. It alludes to human-level capabilities that are better described as AGI (artificial general intelligence)—software capable of any intellectual task humans can do—and no one knows if and when AGI could ever be achieved.

✓ 事实上，机器学习计划在用于优化现有流程时最为有效；这些解决方案类型可以为企业提供最大的投资回报。

✓  In fact, ML initiatives are most effective when used to optimize existing processes; these are the types of solutions that provide the greatest return on investment for businesses.

✓ 将所有 ML 计划纳入“AI”保护伞下，存在过度宣传和误导，导致 ML 业务部署的失败率很高。

✓  Including all ML initiatives under the “AI” umbrella oversells and misleads, contributing to a high failure rate for ML business deployments.

笔记

NOTES

1. 1 . Eric Siegel，“模型很少部署：机器学习领导力的全行业失败”，KD Nuggets，2022 年 1 月 17 日， https:// www .kdnuggets .com /2022 /01 /models -rarely -deployed -industrywide -失败-机器-学习-领导.html。

2. 1. Eric Siegel, “Models Are Rarely Deployed: An Industry-Wide Failure in Machine Learning Leadership,” KD Nuggets, January 17, 2022, https://www.kdnuggets.com/2022/01/models-rarely-deployed-industrywide-failure-machine-learning-leadership.html.

3. 2 . Erik Brynjolfsson 和 Andrew McAfee，“人工智能商业”，hbr.org，2017 年 7 月 18 日， https:// hbr .org /2017 /07 /the -business -of -artificial -intelligence。

4. 2. Erik Brynjolfsson and Andrew McAfee, “The Business of Artificial Intelligence,” hbr.org, July 18, 2017, https://hbr.org/2017/07/the-business-of-artificial-intelligence.

改编自 hbr.org 上发布的内容，2023 年 6 月 2 日（产品#H07NQA）。

Adapted from content posted on hbr.org, June 2, 2023 (product #H07NQA).

本文是作者作为弗吉尼亚大学达顿商学院分析学身体二百周年教授的工作成果。

This article is a product of the author’s work as the Bodily Bicentennial Professor in Analytics at University of Virginia Darden School of Business.

• 亚伯拉罕，马可福音，1-14
• Abraham, Mark, 1–14
• 阿卡尔，奥古兹 A.，95–103
• Acar, Oguz A., 95–103
• 准确性问题, 11 , 12 , 26–27 , 31 , 40 , 72 , 76 , 134 , 138–140
• accuracy issues, 11, 12, 26–27, 31, 40, 72, 76, 134, 138–140
• 适应，连续，110–111
• adaptation, continuous, 110–111
• 行政蠕变，73
• administrative creep, 73
• 非洲联盟，126
• African Union, 126
• 人工智能。请参阅 人工智能 (AI)
• AI. See artificial intelligence (AI)
• 爱彼迎，17 岁
• Airbnb, 17
• 人工智能系统
• AI systems
• 道德发展指南，134–137
• guidelines for ethical development of, 134–137
• 的不可见性和不可理解性，113–114
• invisibility and inscrutability of, 113–114
• 判断何时使用，114
• judging when to use, 114
• 监管，126–127
• regulation of, 126–127
• 保护，124–127
• safeguarding, 124–127
• 的优势，121–122
• strengths of, 121–122
• 值得信赖的原则，133
• trusted principles for, 133
• 山姆奥特曼，95
• Altman, Sam, 95
• 亚马逊，9、21、111–112​​​
• Amazon, 9, 21, 111–112
• 安德森诉 Stability AI 等人，83
• Andersen v. Stability AI et al., 83
• 安迪·沃霍尔基金会，85
• Andy Warhol Foundation, 85
• 安东尼·卡伦，113
• Anthony, Callen, 113
• 吉尔·阿佩尔，81–93
• Appel, Gil, 81–93
• 通用人工智能（AGI ），145–146，149–151
• artificial general intelligence (AGI), 145–146, 149–151
• 人工智能（AI）
• artificial intelligence (AI)
• 另请参阅 生成式人工智能
• See also generative AI
• 的能力，147–148
• capabilities of, 147–148
• 决策者，112–115
• decision-making by, 112–115
• 一般, 145–146 , 149–151
• general, 145–146, 149–151
• 炒作，143–152
• hype about, 143–152
• 在组织中引入，109-112
• introducing, in organizations, 109–112
• 机器学习，144–145，150–152
• machine learning and, 144–145, 150–152
• 狭窄，145–146
• narrow, 145–146
• 人工智能法，126
• Artificial Intelligence Act, 126
• 审计追踪，87–88
• audit trails, 87–88

• 吟游诗人，25 岁
• Bard, 25
• 凯西·巴克斯特，131–142
• Baxter, Kathy, 131–142
• 贝奇基，贝丝 A.，113
• Bechky, Beth A., 113
• 艾米莉·本德尔, 115 , 117
• Bender, Emily, 115, 117
• 最佳实践
• best practices
• 在人工智能决策中，112–115
• in AI decision-making, 112–115
• 为了消除偏见，118–119
• for combating bias, 118–119
• 数据安全，119–120
• for data security, 119–120
• 在组织中引入人工智能，109–112
• for introducing AI in organizations, 109–112
• 缓解风险，115–118
• for risk mitigation, 115–118
• 杰夫·贝索斯，111–112
• Bezos, Jeff, 111–112
• 尼古拉·莫里尼·比安齐诺，33–45
• Bianzino, Nicola Morini, 33–45
• 偏见
• bias
• 算法，39
• algorithmic, 39
• 检查，139–140
• checking for, 139–140
• 担心，132
• concerns about, 132
• 专业知识，53–55
• expertise, 53–55
• 在生成人工智能中，31、72、108、115、117、138–140​​​​​​
• in generative AI, 31, 72, 108, 115, 117, 138–140
• 造成的伤害, 124
• harms caused by, 124
• 减少, 10 , 118–119 , 134–135
• reducing, 10, 118–119, 134–135
• 测试, 11
• testing for, 11
• 兵, 16 , 19 , 113
• Bing, 16, 19, 113
• 双联，52
• bisociation, 52
• 黑匣子，人工智能决策，112–113
• black box, of AI decision-making, 112–113
• 区块链，119
• blockchain, 119
• 样板材料，122
• boilerplate material, 122
• 边界扳手，78
• boundary spanners, 78
• 品牌, 88–90 , 92
• brands, 88–90, 92
• 埃里克·布林约尔松，149
• Brynjolfsson, Erik, 149
• 气泡效果，39
• bubble effects, 39
• 布奥拉姆维尼，乔伊，124
• Buolamwini, Joy, 124
• 企业。查看 组织
• businesses. See organizations
• 生成式人工智能的商业影响，1–16、26、42、107–108、127、132​​​​
• business impacts, of generative AI, 1–16, 26, 42, 107–108, 127, 132
• BuzzFeed，39
• BuzzFeed, 39

• 安·卡沃基安，120
• Cavoukian, Ann, 120
• 改变
• change
• 接受, 110
• acceptance of, 110
• 连续，110–111
• continuous, 110–111
• 聊天GPT, 2 , 49 , 107
• ChatGPT, 2, 49, 107
• 准确性问题, 10 , 76
• accuracy issues, 10, 76
• 禁令, 126
• bans on, 126
• 的能力，19
• capabilities of, 19
• 创造性工作，34
• creative work and, 34
• 使用的数据, 19 , 113–114
• data used by, 19, 113–114
• 促进与65–67的合作
• facilitation of collaboration with, 65–67
• 炒作, 25 , 143
• hype around, 25, 143
• 想法评估者，55–60
• idea evaluation by, 55–60
• 创意产生，51–53
• idea generation with, 51–53
• 使用想法细化，60–62
• idea refinement using, 60–62
• 的影响, 15–16 , 34
• impact of, 15–16, 34
• 学习, 16
• learning by, 16
• 第121章能力的限制
• limits on abilities of, 121
• 受欢迎程度, 72
• popularity of, 72
• 提示工程，37
• prompt engineering for, 37
• 聊天点，8
• ChatSpot, 8
• 首席客户保护官，10-11
• chief customer protection officers, 10–11
• 中国，126–127
• China, 126–127
• 克劳德，25 岁
• Claude, 25
• 共同创作者，92
• co-creators, 92
• 代码
• code
• 人工智能辅助创作, 35 , 36
• AI-assisted creation of, 35, 36
• 第 3–5、8、123代
• generation of, 3–5, 8, 123
• 法典，123
• Codex, 123
• 德文·科尔杜威，145
• Coldewey, Devin, 145
• 合作
• collaboration
• 人工智能，109–112
• with AI, 109–112
• 促进，62–67
• facilitation of, 62–67
• 集体[i]，123
• Collective[i], 123
• 竞争, 38
• competition, 38
• 机密信息，91
• confidential information, 91
• 内容
• content
• 策展，39-40
• curation of, 39–40
• 洪水, 48
• flood of, 48
• 适度，41
• moderation, 41
• 无限，38
• unlimited, 38
• 内容创作
• content creation
• 更改为，38–39
• changes to, 38–39
• 创意，33–45
• creative, 33–45
• 通过生成人工智能，37–39 , 42
• by generative AI, 37–39, 42
• 人类，40–41
• by humans, 40–41
• 个性化，39
• personalization of, 39
• 内容创作者
• content creators
• 数据集的构建，91–92
• building of datasets by, 91–92
• 位移, 92
• displacement of, 92
• 知识产权，87–92
• intellectual property of, 87–92
• 持续适应/变化，110–111
• continuous adaptation/change, 110–111
• 合同，90–91
• contracts, 90–91
• 版权所有, 11 , 84 , 85
• copyright, 11, 84, 85
• 侵犯版权, 38 , 82 , 83 , 84
• copyright infringement, 38, 82, 83, 84
• 创作
• creative work
• 生成式人工智能，33–45
• generative AI and, 33–45
• 减少进入壁垒，36
• reduced barriers to entry in, 36
• 创造力, 34
• creativity, 34
• 人工智能增强，47–69
• AI-augmented, 47–69
• 竞争力，41
• competitiveness and, 41
• 评估57–60的尺寸
• evaluating dimensions of, 57–60
• 人类、37、40–41、43–44、47–69​​​​​
• human, 37, 40–41, 43–44, 47–69
• 机器垄断，37–39
• machine monopolization of, 37–39
• 创造者经济，33–34
• creator economy, 33–34
• 众包, 48 , 62
• crowdsourcing, 48, 62
• 策展, 39–40 , 116
• curation, 39–40, 116
• 顾客
• customers
• 数据来自, 10 , 20–21 , 119–120 , 137–138
• data from, 10, 20–21, 119–120, 137–138
• 最终目标，7-8
• end goals of, 7–8
• 互动, 2–4 , 74–75
• interactions with, 2–4, 74–75
• 自助服务，78–79
• self-service by, 78–79
• 客户服务，人工智能，35
• customer service, AI in, 35
• CVS，9-10
• CVS, 9–10

• 达尔-E, 2 , 35 , 81 , 102–103
• DALL-E, 2, 35, 81, 102–103
• 数据
• data
• 访问, 21
• access to, 21
• 偏差, 10 , 11 , 118–119 , 138
• biases in, 10, 11, 118–119, 138
• 聚集在一起，5-6
• bringing together, 5–6
• 清洁, 5
• cleansing, 5
• 同意使用，135
• consent for using, 135
• 策展, 39–40 , 116
• curation, 39–40, 116
• 客户, 10 , 20–21 , 119–120 , 137–138
• customer, 10, 20–21, 119–120, 137–138
• 文档，117–118
• documentation of, 117–118
• 第一方，137–138
• first-party, 137–138
• 新鲜度, 138
• freshness of, 138
• 治理, 5
• governance, 5
• 管理，10-12
• management of, 10–12
• 元数据，139
• metadata, 139
• 开源，135
• open-source, 135
• 侵犯隐私，119–120
• privacy violations, 119–120
• 处理，121–122
• processing, 121–122
• 代表性，118–119
• representativeness of, 118–119
• 分享, 21
• sharing, 21
• 资料来源, 11 , 20–21 , 116–117 , 135
• sources, 11, 20–21, 116–117, 135
• 调整适当的产出，116-117
• tailoring for appropriate outputs, 116–117
• 第三方, 6 , 9 , 137–138
• third-party, 6, 9, 137–138
• 培训, 82–92 , 113–119 , 135 , 137–138
• training, 82–92, 113–119, 135, 137–138
• 非结构化，72
• unstructured, 72
• AI 使用，17 , 19–21
• used by AI, 17, 19–21
• 零方，137–138
• zero-party, 137–138
• 数据经纪人，137–138
• data brokers, 137–138
• 数据协作软件，5
• data collaboration software, 5
• 数据驱动的学习，17–18
• data-driven learning, 17–18
• 数据湖, 82 , 84 , 88 , 92 , 93
• data lakes, 82, 84, 88, 92, 93
• 数据网络效应，17–18
• data network effects, 17–18
• 数据保护法，126–127
• data protection laws, 126–127
• 决策，人工智能的透明度，112–115
• decision-making, transparency in AI, 112–115
• 大卫·德·克雷默，33–45
• De Cremer, David, 33–45
• 深度赝品，127
• deep fakes, 127
• 深度合成技术，126–127
• deep synthesis technologies, 126–127
• 生成式人工智能的需求，26–30
• demand, for generative AI, 26–30
• 创新民主化, 47–49 , 68
• democratizing innovation, 47–49, 68
• 衍生作品, 84 , 85 , 88 , 89
• derivative works, 84, 85, 88, 89
• 设计理念，快速生产，49–53
• design concepts, rapid production of, 49–53
• 设计固定，53
• design fixation, 53
• 道格·迪茨，100
• Dietz, Doug, 100
• 数字助理，78–79
• digital assistants, 78–79
• 数字思维, 108 , 110 , 123
• digital mindsets, 108, 110, 123
• 披露，90–91
• disclosures, 90–91
• 发散性思维，49–53
• divergent thinking, 49–53
• 文档
• documentation
• 人工智能开发，124–125
• of AI development, 124–125
• 数据，117–118
• of data, 117–118
• 领域专业知识，48
• domain expertise, 48
• 下拉菜单，2
• drop-down menus, 2

• Eapen，Tojin T.，47–69
• Eapen, Tojin T., 47–69
• 电子商务，79
• e-commerce, 79
• 埃德尔曼，大卫·C.，1–14
• Edelman, David C., 1–14
• 阿尔伯特·爱因斯坦，44 岁
• Einstein, Albert, 44
• 爱因斯坦 GPT，71
• Einstein GPT, 71
• 安装效应，53
• Einstellung effect, 53
• 雇员
• employees
• 人工智能协助，139
• AI to assist, 139
• 来自140 的反馈
• feedback from, 140
• 快捷方式，121–122
• shortcuts by, 121–122
• 赋权，135–136
• empowerment, 135–136
• 能源消耗，136–137
• energy consumption, 136–137
• 环境影响, 116 , 136–137
• environmental impacts, 116, 136–137
• 错误, 26–27 , 31 , 40 , 72 , 76 , 115 , 132 , 134 , 138–140
• errors, 26–27, 31, 40, 72, 76, 115, 132, 134, 138–140
• 道德问题, 108 , 125 , 132 , 134–137 , 139
• ethical issues, 108, 125, 132, 134–137, 139
• 道德咨询委员会，140
• ethics advisory councils, 140
• 道德监督机构，125
• ethics watchdogs, 125
• 埃西, 17 , 18
• Etsy, 17, 18
• 欧盟，126
• European Union, 126
• 评估过载，48
• evaluation overload, 48
• 专业知识, 48
• expertise, 48
• 专业知识偏差，53–55
• expertise bias, 53–55
• 埃克森瓦尔迪兹漏油事件，98
• Exxon Valdez oil spill, 98

• 面部识别技术, 124 , 126
• facial recognition technologies, 124, 126
• 公平性，10–11
• fairness, 10–11
• 合理使用原则, 84 , 85
• fair use doctrine, 84, 85
• 本·法尔克，33–45
• Falk, Ben, 33–45
• 谎言，115
• falsehoods, 115
• 法亚德，安妮劳尔，113
• Fayard, Anne-Laure, 113
• 法耶兹，乌萨马，149
• Fayyad, Usama, 149
• 反馈, 19–20 , 140
• feedback, 19–20, 140
• 丹尼尔·J·芬肯施塔特，47–69
• Finkenstadt, Daniel J., 47–69
• 第一方数据，137–138
• first-party data, 137–138
• “五个为什么”技术，98
• “Five Whys” technique, 98
• 翻转板, 33
• Flipboard, 33
• 民谣，乔什，47–69
• Folk, Josh, 47–69
• 功能分解，99–100
• functional decomposition, 99–100
• 功能固定性，53
• functional fixedness, 53

• 格布鲁、蒂姆尼特, 115 , 124 , 125
• Gebru, Timnit, 115, 124, 125
• GE 医疗保健，100–101
• GE Healthcare, 100–101
• 生成式人工智能
• generative AI
• 另请参阅 人工智能 (AI)
• See also artificial intelligence (AI)
• 适应，1-14
• adapting to, 1–14
• 优点, 68
• advantages of, 68
• 作为助理，135–136
• as assistant, 135–136
• 17、19–21的数据需求
• data needs of, 17, 19–21
• 发散思维，49–53
• divergent thinking and, 49–53
• 反馈，19-20
• feedback for, 19–20
• 道德发展指南，134–137
• guidelines for ethical development of, 134–137
• 人类创造力，47–69
• human creativity and, 47–69
• 人类监督，138–139
• human oversight of, 138–139
• 炒作, 3 , 143–152
• hype about, 3, 143–152
• 1–16、26、42、107–108、127、132的影响​​​​​​​​​
• impacts of, 1–16, 26, 42, 107–108, 127, 132
• 积分，137–140
• integrating, 137–140
• 知识产权问题，81–93
• intellectual property issues with, 81–93
• 网络效应，15–23
• network effects and, 15–23
• 挑选项目，25-32
• picking project for, 25–32
• 131的潜力
• potentials of, 131
• 准备，41–43
• preparing for, 41–43
• 监管，126–127
• regulation of, 126–127
• 抵抗力, 68
• resistance to, 68
• 风险管理，131–142
• risk management, 131–142
• 风险来自, 10–12 , 26–27 , 31
• risks from, 10–12, 26–27, 31
• 在销售组织中，71–80
• in sales organizations, 71–80
• 的优点和缺点，121-122
• strengths and weaknesses of, 121–122
• 测试和监督，139–140
• testing and oversight of, 139–140
• 负责任地使用，107–130
• using responsibly, 107–130
• 错误回答，10
• wrong responses from, 10
• 盖蒂，84 岁
• Getty, 84
• GitHub 副驾驶，35 岁，36 岁
• GitHub Copilot, 35, 36
• GoFundMe，102–103
• GoFundMe, 102–103
• 林恩·戈德史密斯，85 岁
• Goldsmith, Lynn, 85
• 谷歌、9、16、21、84、115、125​​​​​​​​
• Google, 9, 16, 21, 84, 115, 125
• 谷歌地图, 18 , 20
• Google Maps, 18, 20
• 治理, 41 , 126–127
• governance, 41, 126–127
• GPT-4, 25 , 96
• GPT-4, 25, 96

• 幻觉, 10 , 12 , 138 , 139–140
• hallucinations, 10, 12, 138, 139–140
• 减少伤害，124–127
• harm reduction, 124–127
• 医疗保健，人工智能，35
• health care, AI in, 35
• 帮助改变一切活动，102–103
• Help Changes Everything campaign, 102–103
• 诚实，135
• honesty, 135
• 枢纽点，8
• HubSpot, 8
• 人类创造力, 38 , 43–44 , 68
• human creativity, 38, 43–44, 68
• 人工智能增强，47–69
• AI-augmented, 47–69
• 溢价，40–41
• premium on, 40–41
• 人类监督，122、135–136、138–140​​
• human oversight, 122, 135–136, 138–140
• 人类表现，增强，123
• human performance, augmentation of, 123
• 人工审查，76
• human review, 76
• 人权，126–127
• human rights, 126–127
• 人类验证，27
• human validation, 27

• 马可·扬西蒂，111
• Iansiti, Marco, 111
• 国际商业机器公司，124
• IBM, 124
• 创意竞赛，60
• idea contests, 60
• 创意产生, 27 , 30 , 48–53
• idea generation, 27, 30, 48–53
• 想法
• ideas
• 评估, 55–60
• evaluation of, 55–60
• 细化，60–62
• refinement of, 60–62
• 图像生成器, 35 , 49–53 , 81–82 , 86–87
• image generators, 35, 49–53, 81–82, 86–87
• 另请参阅 生成式人工智能
• See also generative AI
• 不准确, 26–27 , 31 , 40 , 72 , 76 , 115 , 132 , 134 , 138–140
• inaccuracies, 26–27, 31, 40, 72, 76, 115, 132, 134, 138–140
• 信息
• information
• 另请参阅 数据
• See also data
• 机密, 91
• confidential, 91
• 已连接, 8
• connected, 8
• 错误/虚假信息, 39 , 115
• mis/disinformation, 39, 115
• 信息收集，20
• information gathering, 20
• 创新中心, 98 , 99
• InnoCentive, 98, 99
• 创新
• innovation
• 人工智能辅助，35–37
• AI-assisted, 35–37
• 比赛, 48
• contests, 48
• 民主化, 47–49 , 68
• democratizing, 47–49, 68
• 放缓, 38
• slowing of, 38
• 知识产权
• intellectual property
• 生成式人工智能，81–93
• generative AI and, 81–93
• 法律，38
• laws, 38
• 法律问题，82–85
• legal issues, 82–85
• 监督和执行，88-90
• monitoring and enforcement of, 88–90
• 在训练数据中，87–88
• in training data, 87–88
• 智力, 146
• intelligence, 146
• 意大利，126
• Italy, 126

• 耆那教，丁卡人，15–23
• Jain, Dinkar, 15–23
• 创造就业机会，122
• job creation, 122
• 失业, 47 , 122–123
• job losses, 47, 122–123
• 旅程扩展, 8
• journey expansion, 8

• 知识工作者，43
• knowledge workers, 43

• LAION-5B 数据集，87
• LAION-5B dataset, 87
• 拉卡尼，卡里姆河，111
• Lakhani, Karim R., 111
• 大语言模型 (LLM), 25–26 , 42 , 108
• large language models (LLMs), 25–26, 42, 108
• 另请参阅 ChatGPT；生成式人工智能
• See also ChatGPT; generative AI
• 数据，116–118
• data for, 116–118
• 的不可见性和不可理解性，113–114
• invisibility and inscrutability of, 113–114
• 第121章能力的限制
• limits on abilities of, 121
• 风险，115–116
• risks of, 115–116
• 可持续性，136–137
• sustainability of, 136–137
• 诉讼，83–85
• lawsuits, 83–85
• 潜在客户开发，79
• lead generation, 79
• 学习, 29 , 31
• learning, 29, 31
• 企业, 29 , 31
• corporate, 29, 31
• 数据驱动，17–18
• data-driven, 17–18
• 机器, 16–18 , 108 , 143–145 , 147–152
• machine, 16–18, 108, 143–145, 147–152
• 加固, 18
• reinforcement, 18
• 法律问题，与生成人工智能，82–85
• legal issues, with generative AI, 82–85
• 莱文，辛·S.，15–23
• Levine, Sheen S., 15–23
• 责任，132
• liability, 132
• 骆驼, 25
• LLaMA, 25
• 法学硕士。查看 大型语言模型 (LLM)
• LLMs. See large language models (LLMs)
• 萨莉·E·洛里默，71–80
• Lorimer, Sally E., 71–80

• 机器学习, 16 , 108 , 143–145 , 147
• machine learning, 16, 108, 143–145, 147
• 人工智能和, 144–145 , 150–152
• AI and, 144–145, 150–152
• 数据驱动，17–18
• data-driven, 17–18
• 监督，148–149
• supervised, 148–149
• 值，148–150
• value of, 148–150
• 营销, 27 , 29
• marketing, 27, 29
• 莫瑞泰斯皇家美术馆, 82
• Mauritshuis, 82
• 安德鲁·迈克菲，149
• McAfee, Andrew, 149
• 麦克米兰少校，安吉丽娜，115
• McMillan-Major, Angelina, 115
• 元数据，139
• metadata, 139
• 微软、108、115、124、125​​​​​
• Microsoft, 108, 115, 124, 125
• OpenAI 和, 16 , 19–22
• OpenAI and, 16, 19–22
• 销售万岁，71
• Viva Sales, 71
• 中途, 34 , 49–53 , 81 , 86–87 , 107
• Midjourney, 34, 49–53, 81, 86–87, 107
• 错误/虚假信息, 39 , 115
• mis/disinformation, 39, 115
• 米切尔，玛格丽特，115
• Mitchell, Margaret, 115
• 伊森·莫里克，26 岁
• Mollick, Ethan, 26
• 现代艺术博物馆，81
• Museum of Modern Art, 81
• 马斯克，埃隆，115
• Musk, Elon, 115

• 叙事人工智能，5
• narrative AI, 5
• 狭义人工智能，145–146
• narrow AI, 145–146
• 自然语言处理, 36 , 37 , 96 , 147
• natural language processing, 36, 37, 96, 147
• 朱莉安娜·尼尔鲍尔，81–93
• Neelbauer, Juliana, 81–93
• 尼雷·才达尔，107–130
• Neeley, Tsedal, 107–130
• 负面刻板印象，115
• negative stereotypes, 115
• 网络效应，15–23
• network effects, 15–23
• 对于人工智能，17-18
• for AI, 17–18
• 数据，17–18
• data, 17–18
• 定义, 16
• definition of, 16
• 直接, 17
• direct, 17
• 间接，17
• indirect, 17
• 新颖的想法，48
• novel ideas, 48
• 英伟达，108
• NVIDIA, 108

• 本体论，42
• ontology, 42
• 开放人工智能, 8 , 15 , 108 , 113
• OpenAI, 8, 15, 108, 113
• 微软和, 16 , 19–22
• Microsoft and, 16, 19–22
• 开源数据，135
• open-source data, 135
• 运营模式，将人工智能融入其中，111–112
• operating models, building AI into, 111–112
• 运营决策，144
• operational decisions, 144
• 组织结构，111
• organizational structures, 111
• 组织
• organizations
• 危害减少124–127
• harm reduction by, 124–127
• 将生成式人工智能融入，137–140
• integrating generative AI into, 137–140
• 引入人工智能，109-112
• introducing AI in, 109–112
• 风险管理, 90–91 , 131–142
• risk management for, 90–91, 131–142
• 外包，77–78
• outsourcing, 77–78
• 监督, 122 , 135–136 , 138–140
• oversight, 122, 135–136, 138–140

• 专利侵权，82
• patent infringement, 82
• 个性化, 2 , 12
• personalization, 2, 12
• 内容创作，38–39
• of content creation, 38–39
• 预测, 17 , 18 , 121–122 , 144
• predictions, 17, 18, 121–122, 144
• 预测分析，144
• predictive analytics, 144
• 事前分析，115
• premortem analyses, 115
• 隐私设计 (PbD) 框架，120
• privacy by design (PbD) framework, 120
• 隐私问题, 10–11 , 116 , 119–120 , 135
• privacy issues, 10–11, 116, 119–120, 135
• 问题约束设计，101–103
• problem constraint design, 101–103
• 问题分解，99–100
• problem decomposition, 99–100
• 问题诊断，97–98
• problem diagnosis, 97–98
• 问题表述，96–103
• problem formulation, 96–103
• 问题重构，100–101
• problem reframing, 100–101
• 产品开发
• product development
• 专业知识偏见，53–55
• expertise bias in, 53–55
• 促进合作，62-67
• facilitation of collaboration in, 62–67
• 想法评估，55–60
• idea evaluation in, 55–60
• 创意产生，49–53
• idea generation and, 49–53
• 想法细化，60–62
• idea refinement in, 60–62
• 负责人, 133
• responsible, 133
• 生产力, 36 , 78–79 , 121–122
• productivity, 36, 78–79, 121–122
• 产品推荐，74
• product recommendations, 74
• 提示工程, 36–37 , 43 , 95–103
• prompt engineering, 36–37, 43, 95–103

• 马克·拉莫斯，25–32
• Ramos, Marc, 25–32
• 红迪网，116
• Reddit, 116
• 法规, 126–127 , 132
• regulations, 126–127, 132
• 强化学习，18
• reinforcement learning, 18
• 风险委员会，11
• risk committees, 11
• 风险管理，131–142
• risk management, 131–142
• 风险
• risks
• 生成式人工智能, 26–27 , 31 , 124
• of generative AI, 26–27, 31, 124
• 合法，85–91
• legal, 85–91
• 法学硕士数量，115–116
• of LLMs, 115–116
• 缓解, 86–91 , 115–118
• mitigation of, 86–91, 115–118
• 侵犯隐私的情况，119–120
• of privacy violations, 119–120
• 机器人技术，35
• robotics, 35
• 基于规则的系统，147
• rule-based systems, 147
• 规则层，6–7
• rules layer, 6–7

• 安全, 10–11 , 134–135
• safety, 10–11, 134–135
• Salesforce, 71 , 133
• Salesforce, 71, 133
• 销售经理，75
• sales managers, 75
• 销售组织
• sales organization
• 行政渗透，73
• administrative creep in, 73
• 生成式人工智能对71-80的影响
• impact of generative AI in, 71–80
• 销售人员，78–79
• salespeople, 78–79
• 销售技术，73
• sales technology, 73
• 约阿夫·施莱辛格，131–142
• Schlesinger, Yoav, 131–142
• 大卫·A·施韦德尔，81–93
• Schweidel, David A., 81–93
• 搜索成本，39–40
• search costs, 39–40
• 搜索引擎，15–16
• search engines, 15–16
• 安全，10–11
• security, 10–11
• 安全漏洞，116
• security breaches, 116
• 安全风险，132
• security risks, 132
• 沙阿，达米什，8
• Shah, Dharmesh, 8
• 分享经验, 39
• shared experiences, 39
• 阿伦·沙斯特里，71–80
• Shastri, Arun, 71–80
• 快捷方式，121–122
• shortcuts, 121–122
• 埃里克·西格尔，143–152
• Siegel, Eric, 143–152
• 辛哈，普拉巴坎特，71–80
• Sinha, Prabhakant, 71–80
• 软件应用程序，交互，3–4
• software applications, interactions with, 3–4
• 软件接口，2–4
• software interfaces, 2–4
• 语音识别，147
• speech recognition, 147
• 稳定性人工智能，87
• Stability AI, 87
• 稳定扩散, 49–51 , 53–54 , 62–67 , 81 , 84 , 86–87 , 102–103
• Stable Diffusion, 49–51, 53–54, 62–67, 81, 84, 86–87, 102–103
• 利益相关者，140
• stakeholders, 140
• 斯蒂米特, 33
• Steemit, 33
• 结构设计，64–65
• structural design, 64–65
• 子问题，99–100
• subproblems, 99–100
• 子堆栈，33
• Substack, 33
• 监督机器学习，148–149
• supervised machine learning, 148–149
• 可持续性，136–137
• sustainability, 136–137

• 目标参数，6–7
• target parameters, 6–7
• 技术冲击，40
• techlash, 40
• 技术变革
• technological change
• 创造力，34
• creativity and, 34
• 对122-123 人就业的影响
• impact on jobs of, 122–123
• 拉里·特斯勒，148
• Tesler, Larry, 148
• 测试，139–140
• testing, 139–140
• 文本生成器，82
• text generators, 82
• 第三方数据, 6 , 9 , 137–138
• third-party data, 6, 9, 137–138
• 商业外观，89
• trade dress, 89
• 商标侵权, 82–84 , 89–90
• trademark infringement, 82–84, 89–90
• 商业秘密，85
• trade secrets, 85
• 训练数据, 82–92 , 113–119
• training data, 82–92, 113–119
• 同意使用，135
• consent for using, 135
• 零方或第一方，137–138
• zero-party or first-party, 137–138
• 变革性使用, 84 , 85
• transformative use, 84, 85
• 透明度, 10–11 , 87
• transparency, 10–11, 87
• 在人工智能决策中，112–115
• in AI decision-making, 112–115
• 数据集，125
• of data sets, 125
• 使用人工智能，135
• in use of AI, 135
• 到到网, 9
• Tripadvisor, 9
• 三联，52
• trisociation, 52
• 图灵测试，147
• Turing test, 147

• 优步，21 岁
• Uber, 21
• 技能提升，123
• upskilling, 123
• 用例, 27–30 , 32 , 144
• use cases, 27–30, 32, 144
• 用户
• users
• 与62–67之间的合作
• collaboration with and among, 62–67
• 数据来自，135
• data from, 135
• 在创新过程中，47–49
• in innovation process, 47–49

• 文卡塔斯瓦米，洛克什，47–69
• Venkataswamy, Lokesh, 47–69
• 病毒内容，39
• viral content, 39
• 销售万岁，71
• Viva Sales, 71
• 语音转文本算法，36–37
• voice-to-text algorithms, 36–37
• 埃里克·冯·希佩尔，47–48
• von Hippel, Eric, 47–48

• 看门狗，125
• watchdogs, 125
• 位智应用程序，21
• Waze app, 21
• 和族人群，98
• Wazoku Crowd, 98
• 维基百科，116
• Wikipedia, 116
• 工作分解结构，100
• work breakdown structure, 100
• 史蒂夫·沃兹尼亚克，115
• Wozniak, Steve, 115

• 马克·扎奥·桑德斯，25–32
• Zao-Sanders, Marc, 25–32
• 零方数据，137–138
• zero-party data, 137–138